The Impact of Innovation Assumptions on Volatility Forecasting During the Financial Tsunami

• Main Authors: LI, CHIEN-TSUNG, 李建宗
• Other Authors: YANG, JACK J.W.
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/a8q65u

博士 === 國立雲林科技大學 === 財務金融系 === 105 === The research used five GARCH models and six residual distribution functions to study the volatility forecasts of Taiwan, Shanghai, and Hong Kong stock price indexes during the financial crisis. The effects of error distribution function and trading volume on the volatility forecasting are studied in this research. The squared returns of five minutes intraday data were used as the volatility proxy variables. The Poon and Granger (2003) and Brownlees (2011) rolling window estimation structure method is used as the estimation method of this study. Following Patton (2011), the loss functions are estimated by a combination of imperfect fluctuations in the proxy variables and the robust loss function MSE and QLIKE. QLIKE is not affected by the extreme value of a tail. Therefore, the main loss function selected in this research was mainly from QLIKE with MSE function as the supplement. The paired comparison method and multi-model comparison method were used in the comparisons of loss functions, in which the former employed the Giacomini and White (2006) test (G-W test) and the latter employed the model confidence set (MCS) by Hansen et al. (2011). In the paired comparison between skewed residual distribution and standard residual distribution of GARCH models, both with and without the volume time series in the model, the error distribution functions must be standard when the volatility forecasting performances Taiwan and Hong Kong stock markets are statistically significantly better. Therefore, when using different GARCH models for volatility prediction, it is necessary to consider the variations in different error distribution functions during the global financial crisis. In the case of Taiwan stock market models with trading volume in the skewed distribution, except the EGARCH model, the results were all out of the Model Confidence Set. Different from Taiwan, in the case of Shanghai stock market models with trading volume in the skewed distribution, except the EGARCH model, the results were all in Model Confidence Set. Except EGARCH model, adding trading volume in the models could improve the volatility forecasting in the Hong Kong stock market. Therefore, in the comparison of the volatility predictions of skewed and standard error distributions, there is no consistent conclusion that skewed distribution is better in prediction. The GARCH-type models with trading volume have poor performance of out-of-sample volatility forecasting in Taiwan and Shanghai stock market. However, models with trading volume can improve the volatility forecasting in Hong Kong stock market.