Interplay between distributional and temporal dependence An empirical study with high-frequency asset returns

The recent popularity of copulas in the analysis and modelling of multivariate financial time series arises from several applications in the financial sector. This paper surveys the most important techniques of modelling and measuring distributional dependence with a view towards financial applications such as pricing and hedging financial instruments and portfolio risk management. The term distributional dependence refers to the (contemporaneous) dependence among multiple time series. The majority of results of the existing statistical literature on copulas assumes i.i.d. data. However, real financial time series incorporate temporal dependence such as volatility clustering or seasonality. Moreover, common filtering techniques, for example (G)ARCH filtering, usually also lead to a rejection of the i.i.d. hypothesis due to model misidentification. In this paper we investigate the sensitivity of (distributional) dependence measures with respect to various filtering techniques utilizing an IBM-GM high-frequency data set. The main focus will be on the distributional dependence of extreme events, which is important for risk management. Our results show that filtering techniques crucially affect the distributional dependence structure.