Skip to content
All library documents

Using Range-Based Volatility Estimators in ARCH Forecasting Models

Article Quant Q&A · Author: flxh
Read the original

Summary

The document asks whether range-based estimators such as Garman–Klass can be combined with EWMA or ARCH-family models, and whether a dynamic volatility model is still needed when using such estimators. The response points to research that incorporates alternatives to squared returns into ARCH-style forecasting, including studies using Parkinson and Garman–Klass estimators.

The cited work compares model variants using different forecast error measures, and describes a range-based GARCH approach using Parkinson estimates. This supports treating estimator choice and volatility dynamics as questions that can be evaluated together. However, the document reports no numerical findings or recommendation about which estimator performs best, and gives no detailed implementation procedure. Its examples also do not establish that the methods are appropriate for a couple-minute intraday horizon; that setting would require separate evaluation.

Key ideas

  • Squared returns are not the only input estimator considered for ARCH-family volatility models.
  • Research cited in the document compares Parkinson and Garman–Klass estimators within GARCH forecasting.
  • Forecast performance can be compared using different error measures.
  • The document does not establish a best estimator or validate the approach for very short intraday horizons.

Tags

VolatilityStatisticsBacktesting
Full text
# Combine EWMA or ARCH model with estimator other than squared returns


# Combine EWMA or ARCH model with estimator other than squared returns












Currently I use the EWMA model with the squared logarithmic returns as proxy estimator for the volatility, in order to forecast the volatility one step ahead in an intraday scenario (time frame is a couple of minutes)

However I read and observed the squared returns as volatility estimator has its limitations. So now I want to use a more sophisticated estimator such as Garman-Klass.

My question is:

- Is possible and moreover sensible to combine an estimator such as Garman-Klass with a volatility model like EWMA or any of the ARCH family?

- Or do I even need a volatility model like EWMA or *ARCH when I use these estimator (i.e Garman-Klass) in order to forecast the volatility ?

## Answer by flxh (score 1, accepted)

https://quant.stackexchange.com/a/30454

There are papers about improving *ARCH models by using other estimators than the classic squared returns estimator.

Here are some links:

In the paper How Useful are the Various Volatility Estimators for Improving GARCH-based Volatility Forecasts? Evidence from the Nasdaq - 100 Stock Index for example the researches compare different estimators such as Parkinson, Garman-Klass used in the GARCH model. They provide a measure of how useful these estimators are, by comparing them with different error measures.

In A Range-Based GARCH Model for Forecasting Volatility by Dennis S. Mapa different GARCH models such as Garch(1,2) EGarch and the like are upgraded by using the Parkinson estimator. And then again they are all compared by different error measures.

Both were very interesting reads for me.

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.