A common approach to the estimation of copula-based models is the method of inference functions for margins (IFM). However, IFM is subject to small-sample bias. This paper proposes to estimate Copula-MGARCH models by applying maximization by parts (MBP), a multi-step optimization algorithm. The results show that, although the inland stock markets and the Hong Kong stock markets, London stock markets and the Hong Kong stock markets as well as the inland stock markets and the London stock markets, have weak correlations, but since reunification, the three correlation parameters are all improved, and the correlation between inland and Hong Kong stock markets has increased remarkably. In addition, as another conclusion of this paper, IFM overestimates the conditional covariance matrices compared with MBP. Meanwhile, MBP estimates have much better performance.
展开▼