Bayesian Analysis of a Triple-Threshold GARCH Model with Application in Chinese Stock Market

We construct one triple-threshold GARCH model to analyze the asymmetric response of mean and conditional volatility. In parameter estimation, we apply Griddy-Gibbs sampling method, which require less work in selection of starting values and pre-run. As we apply this model in Chinese stock market, we find that 12-days-average return plays an important role in defining different regimes. While the down regime is characterized by negative 12-days-average return, the up regime has positive 12-days-average return. The conditional mean responds differently between down and up regime. In down regime, the return at date t is affected negatively by lag 2 negative return, while in up regime the return responds significantly to both positive and negative lag 1 past return. Moreover, our model shows that volatility reacts asymmetrically to positive and negative innovations, and this asymmetric reaction varies between down and up regimes. In down regime, volatility becomes more volatile when negative innovation impacts the market than when positive one does, while in up regime positive innovation leads to more volatile market than negative one.