| The joint distributional characteristics of asset returns are of central interest in financial economics. The notion of covariance or correlation between asset returns has played an important role in asset-pricing models and portfolio selection problems. More recently, estimating the full shape of multivariate distribution is rapidly becoming a very active and important area of research. Following the success of the univariate GARCH model, many researchers have extended the univariate GARCH model to multivariate dimension. While considerable progress has been made in modeling conditional covariance in the multivariate GARCH model, there has been little work on the multivariate conditional distribution function. Unlike the univariate GARCH field where many authors have tried to introduce flexible distributions to capture skewness and leptokurtosis in financial asset returns, most academic studies on the multivariate GARCH model have relied on multivariate normal distribution. The purpose of this dissertation is to introduce into the multivariate GARCH model a flexible multivariate parametric distribution: the SU-normal distribution originally introduced by Johnson (1949). The SU -normal distribution is so flexible as to capture a wide range of asymmetry and excess kurtosis. We apply the multivariate S U-normal distribution to modeling time varying dependence of asset returns. Testing the model using daily observations of exchange rates, we find that the new model captures the asymmetry and excess kurtosis much better than the normal distribution model. |
