A hybrid modeling approach for forecasting the volatility of REITs index in US market

Nonlinear estimation is widely accepted by many studies that analyze the financial market, and neural network is one of the effective methods to predict the volatility of market return, especially the US Real Estate Investment Trusts market (REITs). Unfortunately, many of these studies fail to consider alternative techniques of data mining, the relevance of input variables, as well as the performance of modeling. This paper introduce the informatics techniques to select the most relevant input variables for the REITs market, and evaluate the predictive relationship between NAREIT REITs index and numerous financial and economic variables. In this study, we implement the hybrid models which incorporate a series of GARCH family model and artificial neural network (ANN) to examine their ability to provide an effective forecast of future volatility of US REITs market. Our results suggest that Exponential general Autoregressive Conditional Heteroskedastic-ity (EGARCH) model has the highest predict power to the volatility of NAREITs index. Furthermore, the hybrid mode ANN-EGARCH model perform a outstanding predictive power for the in-sample forecasting.