Bjorn Baars

Heavy and Realized (E)garch Models Bjorn Baars

This book investigates the out-of-sample performance of several models that predict unobserved conditional variance. The models that are considered are the HEAVY, RealGARCH(1,1) and the RealEGARCH(1,1) model. These models are also extended, using the squared daily return as extra regressor and adding an indicator function for negative returns multiplied with the realized measure. With these models, forecasts are made and compared with two benchmark models, being the GARCH(1,1) model and the HAR-3 model. The loss function that is used to compare these models is the QLIKE loss function, with the squared daily returns, realized variance and realized kernel as a proxy. The data that are considered, are the indices of the FTSE100, DAX30, CAC40, AEX, SSMI, IBEX35 and the EUROSTOXX50 from January 2000 to March 2014.