Markov Switching Asymmetric GARCH Model and Artificial Neural Networks: Enhancing the volatility forecasting for S&P 500 Index

Financial time series exhibit different stylized facts, namely, asymmetry and nonlinearity, which require a particular specification to capture market volatility behavior. This paper suggests Back-propagation neural networks (BPNN) to improve the forecast accuracy for the S&P 500 returns volatility. The estimated volatility based on the Markov-Switching asymmetric GJR-GARCH (MS GJR-GARCH) model and the VIX index (i.e., Volatility index) series are used respectively as input and output of our neural networks model. The results reveal that the neural networks have succeeded in enhancing the forecast accuracy of the MS GJR-GARCH model according to Mean Squared Error (MSE) and Mean Absolute Error (MAE) criteria.