Estimation of the XU100 Index Return Volatility with the Integration of GARCH Family Models and ANN

To avoid market risk, besides portfolio diversification, investors are trying to model the volatility in the best way to reduce uncertainty. The most commonly used methods to model volatility are Autoregressive Conditional Heteroscedasticity family models. However, some studies published in recent years show that semi-parametric hybrid models which consist of the integration of Autoregressive Conditional Heteroscedasticity family models with the Artificial Neural Networks algorithm perform better than these simple models. In this study, the volatility of the Borsa İstanbul 100 price index return is estimated by the aforementioned simple and hybrid models, and the success of the hybrid models is compared with the success of their components. Two different distributions - Normal Distribution and Generalized Error Distribution - are assumed through the estimation process. Comparisons are performed using Mean Squared Error and Mean Absolute Deviation criteria. According to both criteria, the hybrid model composed of Exponential Generalized Autoregressive Conditional Heteroscedasticity - Artificial Neural Networks performed the best. In line with these findings, it is recommended to evaluate the advantages of hybrid models in dynamic risk analysis of financial instruments.