The Role Of Skewness In Garch Modeling

Libing Fang,Bingshen Kuo,Yong Zeng
2009-01-01
Abstract:The importance of involving skewness into GARCH modeling is documented by simulation and empirical evidence. Since GJR-GARCH is widely used, we make simulation by Monte Carlo method based on such model. The simulation results show that if the innovations are generated from a skewed distribution, the quasi-maximum-likelihood estimates (MLE) are quite probably different from the true values when there is NO skewness parameter in the assumed model. On the contrary, whether or not the symmetry is satisfied, we can always involve a skewness parameter into model, because MLE will only deviate from the true values with a small probability, though exceeding our significance level slightly. In addition, we employ daily index returns of the Chinese stock market to examine the accuracy of the out-of-sample volatility prediction. Various loss functions also confirm the necessity of a skewness parameter for GARCH modeling. Besides, when significant asymmetry is existent, the model conditional on Normality may not always be outperformed by the other more flexible distributions, such as Student-t, which parameterize the stylized fact of thick tail and leptokurtic but in the absence of a skewness parameter. Finally, the above conclusion is robust under the EGARCH and APARCH framework.
What problem does this paper attempt to address?