Nonlinearity and forecast performance of realized volatility

Abstract This study examines whether accounting for the nonlinearity of realized volatility leads to better forecast performance. We propose a new realized volatility forecasting model that considers nonlinearities without the assumption of a particular nonlinear model. The proposed model uses the Taylor series approximation method to account for nonlinearities. We applied it to the realized volatility of representative stock indices from the U.S., Japan, the U.K., and China and observed their in-sample nonlinearities. Additionally, we evaluate out-of-sample forecast performance. The empirical results show that realized volatility has nonlinearity, and the proposed models exhibit better forecast performance than standard models.