Augmented HAR

Abstract

This paper proposes a novel time-series data augmentation algorithm for stock realized volatility forecasting coined as Augmented HAR (AHAR). The findings of this study show that the employment of the AHAR algorithm with artificial neural network (ANN) models statistically significantly enhances the forecast accuracy of these models for newly introduced stocks lacking more than seven years of market data. The AHAR algorithm allows the ANN models to more properly learn nonlinear patterns of the stock realized vo

Keywords: Neural Networks, Realized Volatility Forecasting, Time-Series Data Augmentation.

JEL Classification: C45, C53, C15