Volatility Forecasting Models and High-Frequency Data Effects

Summary

The discussion names several approaches to forecasting asset volatility from returns. It mentions stochastic volatility and GARCH families, including FIGARCH and component GARCH variants for high-frequency settings, and highlights the HAR realized volatility model as a practical choice for combining simplicity with accuracy. The question also raises forecasting horizons from daily to monthly and asks about stocks and bonds, but the answers do not compare model performance by asset class or horizon.

For very high-frequency observations, the responses emphasize that market microstructure effects cannot simply be ignored when estimating volatility. Intraday seasonality and serial dependence are cited as features that need attention. The exchange provides pointers rather than implementation details, data requirements, or empirical comparisons, so it serves as an introductory map of model families rather than a complete forecasting workflow.

Key ideas

• Stochastic volatility and GARCH models forecast volatility using asset returns.
• FIGARCH and component GARCH variants are mentioned for high-frequency data.
• HAR realized volatility is presented as a relatively simple and accurate model option.
• High-frequency volatility estimation must account for market microstructure, intraday seasonality, and autocorrelation.

Tags

# Forecast of volatility


# Forecast of volatility












What are the well known methods for forecasting (daily - weekly - monthly) volatility of a stock price? How about a bond price?

Let's say I have in my disposition the price time series at a very high frequency. How should I avoid dealing with the micro structure of the market?

Thanks

## Answer by Alejandro Andrade (score 0, accepted)

https://quant.stackexchange.com/a/22187

For your first question there a 2 types of models: stochastic volatility and GARCH which need as an imput the return of asset. For HF data this models can be modified, I am not familiar with SV models but GARCH type would be a fractional integraded GARCH (FIGARCH) or a multiplicación componentes GARCH (mcsGARCH). There are some stailized fact od HF data that you have to take into acount like diurnal seasonality and a lot of autocorrelation, there are more that you could search un google or search note un this page bacause there are tons of cuestiona about thia topic

## Answer by Alex C (score 1)

https://quant.stackexchange.com/a/22211

My favorite model of volatility is the HAR RV model by Fulvio Corsi. Good combination of simplicity and accuracy.

http://jfec.oxfordjournals.org/content/7/2/174.short?rss=1&ssource=mfc

About the microstructure issue: you cannot avoid dealing with it, it is important if you are computing volatility from very high frequency.

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.