关键词: 偏斜, 条件分布, GARCH, 波动率预测

Abstract: A Monte Carlo simulation is conducted to examine the effect of skewness parameter on the estimated results of GARCH family models. The results show that the maximum-likelihood estimates conditional on symmetric and thick tailed distributions will produce asymptotic bias when standardized disturbing terms are skewed with heavy tail. Employing the dataset from Shanghai stock market and the robust tool of SPA test, we empirically compare the performances of volatility prediction based on 10 members of GARCH family when their conditional distributions are alternatively specified as Normal, Student-t, GED and Skew-t. The results show that: (1) the best volatility prediction is provided by Skew-t for each of the GARCH structures; (2) volatility prediction conditional on Normality can outperform the other more flexible but symmetric distributions, such as Student-t under eight types of GARCH members, and GED for 4 types; (3) the main conclusion are not altered after varying the quantity of the out-of-sample observation or introducing the term of GARCH-m into mean equation.

Key words: Skewness, Conditional Distribution, GARCH, Volatility Prediction