The role of uncertainty in forecasting volatility comovements across stock markets

Earlier studies found that uncertainty is important in forecasting the financial market covolatilities. However, it is not clear how uncertainty affects the covariance matrix dynamics across different market and economic conditions. To fill this gap, we specify the dynamic relationship between stock market covolatilities and uncertainty in a nonlinear framework, and we analyze the relevance of uncertainty measures in anticipating the transition of conditional covariances between different regimes. Specifically, we propose alternative transformations of the realized covariance matrix which we model by means of the Vector Logistic Smooth Transition Autoregressive (VLSTAR) model. Empirical results indicate that uncertainty measures used as transition variables help to detect covolatilities changes; moreover, the VLSTAR exhibits a significantly better forecast performance compared to alternative linear and multivariate GARCH models. Finally, our results show that the evidence on the role of macroeconomic and financial predictors is mixed, depending on the specification of the realized covariance dynamics.