In this paper, we propose a hybrid model, which combines artificial neural
networks (ANN) with GARCH-type models, to improve the volatility forecasting performance of GARCH-type models in Taiwan stock index. The
realized range-based volatility is used as the true volatility proxy in evaluating forecasting performance while adopting statistical loss functions. A VaR-based loss function is employed to evaluate the predictive performances to further show economic benefits of this hybrid model. To control for the data-snooping problem, the superior predictive ability (SPA) test of Hansen (2005) is applied to reveal the statistical significance and ensure obtaining robust results. Our empirical result indicates that using artificial neural networks can indeed improve the GARCH-based volatility forecasting. However, the improvement is
only limited to the statistical evaluation method.