Forecasting Volatility from One-Minute Data with Realized Measures

Summary

The document asks how to forecast volatility when one-minute bar data is available. It outlines several broad approaches: GARCH models, stochastic volatility models, implied volatility, and realized volatility. Since options prices are unavailable, implied volatility is not considered applicable to the questioner’s data.

The proposed workflow is to calculate realized volatility from the high-frequency observations, then apply standard time-series forecasting methods to the resulting series. The document raises this as a possibility rather than presenting a tested model, comparison, or empirical result. It does not specify how to construct the realized measure, choose a forecast horizon, or handle market microstructure noise and other high-frequency data issues, so those choices remain open.

Key ideas

• One-minute observations make realized volatility a candidate measure for volatility analysis.
• The document contrasts GARCH, stochastic volatility, implied volatility, and realized volatility approaches.
• Without options prices, implied volatility is outside the proposed analysis.
• A possible workflow is to forecast a realized volatility series with time-series methods.
• The question leaves model selection and empirical validation unresolved.

Tags

# Modelling volatility for higher frequency data


# Modelling volatility for higher frequency data












I'm doing some academic work on volatility forecasting. I've got 1-minute bar data. It is not clear to me what model is best suited for forecasting volatility when higher frequency data is available.

I understand the following families/classes of volatility models exist:

- (G)ARCH family models

- Stochastic Volatility

- Implied Volatility (not applicable because I don't have options prices)

- Realised volatility

I was wondering, considering I have high frequency data, realised volatility should provide a reasonable approximation. I could potentially calculate the volatility using realised volatility and then use standard time series forecasting methods to forecast this (realised volatility) series?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.