Dynamic Forecasting: Efficacy of Rolling Symmetric and Asymmetric GARCH Models

Volatility structure of a financial time series is an interesting issue for investors. Most of the research related to volatility forecasting appears to be piecemeal, focusing only on some part of the problem. The author has attempted to overcome this constraint by providing a comprehensive approach to volatility modeling, incorporating comparison of alternative models, rolling data windows, best-fit as well as forecast evaluation measures, additional regressors and different error densities. The time-varying volatility of the Jakarta Stock Exchange Composite Index (JCI) was studied in detail using GARCH models based on dollardenominated daily data for a period from January 2009 through April 2015, with and without exogenous triggers. The results showed that the EGARCH model using Student’s t-distribution was the best among the estimated models for forecasting the futures values and variance of JCI. The results also confirmed the inadequacy of the GARCH (1,1) model in effectively modeling the temporal behavior of financial time series’ volatility.