On the predictability of realized volatility using feasible GLS

Abstract

This study deals with the out-of-sample predictability of realized volatility induced by implied volatility using FGLS. The original dataset was collected from Bloomberg and includes price and implied volatility indices from the US, Hong Kong, China, South Korea and India. Prices were then transformed into realized volatility indices. The relation between realized and implied volatility is important insofar as market expectations about future turbulence may affect the investor's behavior in advance. However, there are some features of the financial data which turn problematic the choice of the OLS estimator. These features include endogeneity and persistence of the predictor, and also conditional heteroskedasticity of the predicted innovations. Consequently, OLS becomes biased and inefficient. The FGLS estimator accounts for these characteristics and, therefore, performs better than OLS-based estimators, as indicated by many of our results.