Can Skewed Garch-Type Distributions Improve Volatility Forecasts During Global Financial Crisis?

This paper is related to the work of Patton (2011), who proposed the required robust loss functions MSE and QLIKE for imperfect fluctuations in the proxy variables, as well as the use of GW and MCS test for statistical analysis. In the same volatility model, the use the GW test pairing for comparing volatility forecasts of skewed and non-skewed error distributions. With the exception of EGARCH, the results produce no clear evidence of better prediction by a non-skewed distribution. In the same volatility model, the comparison of six different error distribution functions for volatility forecast showed no consistent result. In addition to the APARCH model with skewed Student-t distribution, the remaining results favored in nonskewed error distribution function for better prediction. In the comparison of all 30 models for forecasting volatility, better prediction models were all based on APARCH with six different error distribution functions. However, with a 90% confidence level, according to MCS tests, they all were included in the set of better volatility prediction models. A return with skewness, leptokurtic, and thick tail does not necessarily have the best performance in volatility prediction in the skewed error distribution