Dependence modelling is often essential in the practice of risk management and capital allocation,particularly in the case of high-dimensional portfolios.In recent years,vine copulas are receiving increasingly attention due to its potential in modeling high dimensional dependence of financial data.Vine copula decomposes a multivariate copula into a product of a series of bivariate copulas,and hence it can be used to capture the asymmetry and heavy-tailedness in a joint multivariate distribution.Nevertheless,its usability comes along with an exponentially increasing complexity in large dimensions.A d dimensional vine copula consists of d(d-1)/2 bivariate copulas and this leads to a large number of parameters for high-dimensional applications.To develop more parsimonious vine models,in this paper we propose a regularization method to reduce the number of parameters.We shrink the bivariate copula parameter estimators by imposing penalty functions such as LASSO and SCAD in maximum likelihood estimation in such a way that the insignificant bivariate dependence diminishes.We call the resulting vine copula “sparse vine copula”.In term of applications,both the normal and the sparse vine copulas are applied to several financial datasets.By using the criterion of BIC,the sparse vine copula models consistently out-perform the normal ones.Comparison between these two models is also conducted in terms of risk measures for certain portfolios.This is a joint work with Dezhao Han and Ken Seng Tan.