Portfolio Risk VaR Measurement Based On Vine-Copula-EVT

In the theoretical model, the paper in the theoretical framework of Vine-Copula Pair-Copula high-dimensional modeling method based on combining extreme value theory EVT, constructed Vine-Copula-EVT model, the various C-Vine and D-Vine structural decomposition under Copula function common form (Gaussian-Copula, T-Copula, Clayton-Copula, Gumbel-Copula, Frank-Copula) to study the market risk multi-asset portfolio, combined with Monte Carlo simulation method to calculate the multi-asset portfolio VaR as risk metrics through VaR forecasting effect Kupiec failure rate testing Vine-Copula-EVT model and predict the effect of the VaR in multivariate Copula obtained VaR, the traditional Monte Carlo simulation, historical simulation method by comparison.In the empirical process, this paper selects the Shanghai Composite Index, the Nikkei 225 Index, the S & P 500 Index and the FTSE 100 Index as the underlying structure of multi-asset investment portfolios to January 1,2011 to December 31,2015 to sample interval, to give a total of four daily return series 1214 set of sample data to conduct empirical research. In the multi-asset portfolio VaR estimate when, we use a two-stage maximum likelihood estimation method. The first step is to build the edges of the distribution function, the first to return series unique characteristics typical financial data peak, thick tail, skewed and asymmetrical, etc., using GARCH classmodel (GARCH-N, GARCH-t, GARCH-SKST, GJR-N, GJR-t and GJR-SKST) were each index return series filter, which in accordance with the selection criteria Akaike AIC optimal GARCH class model, and the corresponding standard residuals. Then apply extreme value theory EVT, tail distribution with GPD (generalized Pareto distribution) model fit within the threshold distribution is estimated by the empirical distribution function to the extreme value distribution for marginal distribution modeling. The second step is the Vine-Copula dependency modeling. For the dependence structure between the four index, select C-Vine vine structure, Pair-Copula function form of Gaussian-Copula, T-Copula, Clayton-Copula, Gumbel-Copula, Frank-Copula in based on AIC criteria are selected to give the corresponding Vine-Copula dependency modeling, computational multi-asset portfolio VaR by Monte Carlo method. After VaR estimated, the need for examining the results of its paper its VaR obtained with other methods (Multivariate Copula-GARCH, Monte Carlo method) VaR together Kupiec test failure rate, compared to the models to accurately estimate VaR.The empirical results obtained in this paper the following points:1, according to AIC and BIC criteria guidelines, GARCH class model, GJR-SKST model excluding the Shanghai Composite Index, the Nikkei 225 index, the FTSE 100 index rate of return, ARMA (1,1)-GJR-SKST typical financial model excluding the fact that the S & P 500 index feature works best.2, according to AIC criterion, C-Vine-Copula is better than D-Vine-Copula.3, according to the Kupiec failure rate test results, obtained in addition to historical simulation VaR is not good for the future of multi-asset portfolio risk prediction, the other in several ways resulting VaR have good prediction, in which C-Vine-Copula-the effect of EVT forecasting model best. The number of days from the failure, the failure of a few days C-Vine-Copula-EVT are desired within a few days, while other models fail a few days there are more than the expected number of days the situation.