Abstract: | A multiplicative error model with time-varying parameters andan error term following a mixture of gamma distributions isintroduced. The model is fitted to the daily realized volatilityseries of deutschemark/dollar and yen/dollar returns and isshown to capture the conditional distribution of these variablesbetter than the commonly used autoregressive fractionally integratedmoving average model. The forecasting performance of the newmodel is found to be, in general, superior to that of the setof volatility models recently considered by Andersen et al.(2003, Econometrica 71, 579625) for the same data. |