Abstract: The structure of high-dimensional dynamic vine copula can overcome ‘dimensional curse’ faced by bivariate Copula and dynamically describe nonlinear dependence between multi-variables, and represents the academic frontier. Five kinds of foreign exchange log-returns, including USD, EUR, JPY, HKD and GBP, are selected to make empirical analysis, Time series are fitted with AR-GJR-GARCH and GPD models. After probability integral transform, high-dimensional dynamic C and D vine copulas are modelled. Then, portfolio VaR sets are got by Monte-Carlo method, tested by UC back testing, and compared to the corresponding static research. The results show that VaR based on high-dimensional dynamic C vine copula performs the best, and marginal risk of USD is considered as the least by VaR decomposition, the more USD the lest risk. The conclusions provide a new model and method to quantify risk, reasonably allocate asset portfolio, and for authorities to regulate risk.

Key words: AR-GJR-GARCH model, high-dimensional dynamic vine copula, portfolio risk of foreign exchange, VaR, UC back testing