| Summary: | 碩士 === 淡江大學 === 財務金融學系碩士班 === 97 === We compare the predictive performance of various GARCH family models and Neural Networks. The models are compared out-of-sample using Taiwan Stock Exchange Capitalization Weighted Stock Index（TAIEX）data. We substitute the Realized Range-Based Volatility for the latent true volatility and choose six statistical loss functions to compare the predictive performance. We also use the forecasting volatilities into Black-Scholes formula to evaluate the theoretical option prices and compare with real option prices. To control for the fact that as the number of models increase, so does the probability of finding superior predictive ability among the collection of models, we implement the Superior Predictive Ability Test of Hansen（2005）.
We find that, for four loss function, Neural Networks nested Q-GARCH model seems dominate. For two VaR based loss function, GJR-GARCH and GARCH models are preferred. For option pricing, Neural Networks nested I-GARCH model, which performs the worst in the six loss function, seems to be the best performer.
|
|---|
