Bayesian influence diagnostics for a multivariate GARCH model

In this paper, we introduce a diagnostic method for identifying influential observations in the multivariate DCC-GARCH model. We employ the Bayesian local influence method by introducing small perturbations to the prior, variance, and data to assess their impact. Subsequently, through simulation studies and empirical analysis, we demonstrate the effectiveness of the Bayesian local influence method for multivariate GARCH models. In the empirical part, a bivariate GARCH model is established using the daily returns of the S&P 500 Index and IBM, and a comparative analysis is conducted to examine the differences in the influential points detected by the Bayesian method and traditional methods.