Exploring extension of HAR volatility prediction

Abstract

This paper proposes the HAR-weighted model, introducing an inverse standard deviation weighting scheme to the HAR-RV framework - a methodological innovation previously unexplored in the volatility forecasting literature. Our approach systematically mitigates the model’s sensitivity to high-volatility periods through variance-adaptive weighting while preserving the interpretability of the original specification. Theoretically, we establish the first formal asymptotic theory for HAR-type estimators under Elastic Net Regularization, resolving important open questions in robust volatility estimation. The model further enhances predictive performance through judiciously designed non-linear transformations. Comprehensive empirical analysis demonstrates consistent outperformance relative to benchmark specifications, which can be applied in the field of Value-at-Risk. This work both advances the methodological frontier of realized volatility modeling and delivers practical improvements for financial risk management.