Time-varying higher moments, economic policy uncertainty and renminbi exchange rate volatility

We investigate the predictive value of time-varying higher moments and economic policy uncertainty (EPU) for renminbi exchange rate volatility. To do so we develop a generalized autoregressive conditional heteroscedasticity (GARCH) mixed data sampling (MIDAS) model with skewness and kurtosis (the GARCH-MIDAS-SK model), which accommodates time-varying non-Gaussianities (higher moments) of the renminbi exchange rate return distribution and allows us to link volatility to EPU. An empirical analysis based on daily USD/CNY exchange rate returns and monthly global EPU index data shows that the GARCH-MIDAS-SK-EPU model, which incorporates time-varying higher moments and global EPU, can yield more accurate out-of-sample renminbi exchange rate volatility forecasts than the various competing models (ie, the GARCH, GARCH-MIDAS, GARCH-MIDAS-EPU and GARCH-MIDAS-SK models). The superior predictive power of the GARCHMIDAS- SK-EPU model is robust to an alternative version of the global EPU index, alternative out-of-sample forecasting windows and local EPU indexes (Chinese EPU and US EPU). Our empirical findings highlight the value of incorporating timevarying higher moments and EPU into forecasts of renminbi exchange rate volatility.