Publication Type

Journal Article

Version

acceptedVersion

Publication Date

3-2026

Abstract

Standard realized volatility (RV) measures estimate the latent volatility of an asset price using high frequency data with no reference to how or where the estimate will subsequently be used. This paper presents methods for “tailoring” the estimate of volatility to the application in which it will be used. For example, if the volatility measure will be used in a specific parametric forecasting model, it may be possible to exploit that knowledge to construct a better measure of volatility. We use methods from machine learning to estimate optimal “bespoke” RVs for heterogeneous autoregressive (HAR) and GARCH-X forecasting applications. We apply the methods to 886 U.S. stock returns and find that bespoke RVs significantly improve out-of-sample forecast performance. We find that, across a variety of volatility models, the bespoke RV places more weight on data from the end of the trade day, and that the optimal bespoke weights can be well-approximated by a simple parametric function.

Keywords

Volatility forecasting, Machine learning, High frequency data

Discipline

Econometrics

Research Areas

Econometrics

Publication

Journal of Econometrics

Volume

254

First Page

1

Last Page

19

ISSN

0304-4076

Identifier

10.1016/j.jeconom.2025.106122

Publisher

Elsevier

Copyright Owner and License

Authors

Additional URL

https://doi.org/10.1016/j.jeconom.2025.106122

Download

Included in

Econometrics Commons

Share

COinS