Some Forecast Asymmetric GARCH Models for Distributions with Heavy Tails

Abstract

Crude oil prices are inuenced by a number of factors that are far beyond the traditional supply and demand dynamics such as West Texas Intermediate (WTI), Brent and Dubai. The high frequency crude oil data exhibit non-constant variance. This paper models and forecasts the exhibited uctuations via asymmetric GARCH models with the three commonly used error distributions: Student's t distribution, normal distribution and generalized error distribution (GED). The Maximum Likelihood Estimation (MLE) approach is used in the estimation of the asymmetric GARCH family models. The analysis shows that volatility estimates given by the exponential generalized autoregressive conditional heteroskedasticity (EGARCH) model exhibit generally lower forecast errors in returns of WTI oil spot price while the asymmetric power autoregressive conditional heteroskedasticity (APARCH) model exhibits lower forecast
errors in returns of Brent oil spot price, therefore they are more accurate than the estimates given by the other asymmetric GARCH models in each returns. The results obtained from the volatility forecasts seem to be useful to oil future traders and policy makers who need to perceive "apriori" the effects of news on return volatilities before executing their trading, investments and political strategies for the economic wellbeing of the country.