Abstract: The return of financial asset usually has a characteristic of fluctuation clustering with sharp peaks and fat tails, not complying with the normal distribution. Therefore, the nonlinear correlation should be considered when measuring the risk of an investment portfolio. In this respect, copula functions provide a fairly new approach for connecting the marginal distributions of nonlinear series in the high-dimensional risk assessment of portfolio. However, it is noteworthy that two challenging problems exist in this field:one is how to choose or construct an appropriate copula function, the other is how to estimate model parameters. To address these issues, a novel M-Copula-SV-t model is proposed in this paper. Specifically, SV-t model is first employed to fit the marginal distributions of financial time series, where MCMC method with Gibbs sampling are used to estimate marginal parameters; then an M-Copula function consisted of linearly combined Archimedean Copulas is designed to jointly connect these marginals. where joint model parameters are estimated by MLE and BFGS algorithm; afterwards Monte Carlo technique is adopted to simulate optimal portfolios under minimal values of VaR and CVaR. The model feasibility and effectiveness is fruther vertified by taking an example of four exchange rates, where the empirical results indicate that our mixture modeling outperforms other individual Archimedean Copula modeling in dealing with the issue of dimensionality curse, and capturing asymmetry and tailed fatness of portfolio analysis. Therefore, our proposed model contributes to the literature of intra-market portfolio management, and provides valuable suggestions for international investors with respect to short-term decisions.

Key words: mixture copula, SV-t model, CVaR, risk measurement of portfolios, monte carlo simulation