Linear vs. nonlinear GARCH models for RTS index returns

This paper compares the predictive abilities of linear vs. nonlinear GARCH models for conditional volatility for the RTSI returns. Using daily data on the RTSI over past 10 years we estimated the models, obtained predicted values for different horizons, and compared the predictive abilities according to selected criteria. Nonlinear models were created to capture stylized facts about time series, but the quality of the obtained forecasts is sometimes questionable. The results of this study complement the results of other authors — namely, that the nonlinear GARCH models for the conditional volatility show the best results. This result might be due to the fact that nonlinear models show better results than linear ones on relatively short horizons, while on the longer ones their predictive ability might be weakened.