Forecasting Volatility with Encompassing and Regime Dependent GARCH Models

This paper develops the Regime Dependent Generalized Autoregressive Conditional Heteroskedasticity (RD-GARCH) model and applies it to a daily index of returns on U.S. equities. The RD-GARCH model is different from previous models in that it combines Hentschel's single specification that nests several of the more popular extensions to the GARCH model with a general approach that allows model parameters to vary across periods of differing unconditional volatility. The out-of-sample forecasting performance of the RD-GARCH methodology is found to be superior to a number of alternative models. The sensitivity of forecast accuracy to the distributional assumption (normal, Student-t, generalized error) and to the length of model calibration period is also evaluated