Skip to content
All library documents

Using GARCH-MIDAS with Intraday Data and Daily Covariates

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

Summary

The discussion considers whether GARCH-MIDAS, commonly applied to daily returns, can model one-minute returns while incorporating daily exogenous variables. It suggests adapting the model to include an intraday volatility component, with a multiplicative component GARCH approach as a possible way to represent intraday seasonality.

The replies also note that intraday volatility clusters and that liquidity measures such as bid-ask spreads, traded volume, and top-of-book depth may help explain it. One cited study is described as finding improved volatility modeling from traded volume, particularly for US stocks, with some additional gain from the other liquidity variables. The exchange gives suggestions rather than a complete specification or validation of an intraday GARCH-MIDAS model; implementation and empirical performance would need to be assessed for the target data and market.

Key ideas

  • GARCH-MIDAS may be adapted to intraday returns by adding an intraday volatility component.
  • Intraday seasonality is a modeling consideration when working with high-frequency returns.
  • Bid-ask spreads, traded volume, and first-level order-book depth may help model intraday volatility.
  • The cited evidence reports benefits from volume and smaller additional gains from other liquidity variables.

Tags

VolatilityHigh-frequency tradingMarket microstructureEquitiesStatistics
Full text
# Can you use GARCH-MIDAS for intraday data?


# Can you use GARCH-MIDAS for intraday data?












I'm working on a project to forecast volatility and I'm using intraday data (1 min). I want to include exogenous variables to the model that have daily frequency. I was wondering if GARCH-MIDAS can be used for this? The papers I have read on this model use daily price data and the R-package description (mfGARCH) also says

> [...] The GARC-HMIDAS model decomposes the conditional variance of (daily) stock returns into a shortand long-term component, where the latter may depend on an exogenous covariate sampled at a lower frequency.

Thanks!

## Answer by teekanne (score 2)

https://quant.stackexchange.com/a/44409

I guess its possible if you employ some kind of GARCH with an intraday component. In general, it should not be too difficult to alter my R-package mfGARCH for estimating it. Maybe

http://www.unstarched.net/2013/03/20/high-frequency-garch-the-multiplicative-component-garch-mcsgarch-model/

could be a start for modeling intraday seasonality.

Best, Onno

## Answer by lehalle (score 1)

https://quant.stackexchange.com/a/48696

It is a good idea indeed to use GARCH for intraday volatility because it is as clustered as daily volatility. Moreover, if you want to account for autocorrelations, you should consider using other variables like the bid-ask spread, the traded volume and the volume of the book at first limits. It is done in Endogeneous Dynamics of Intraday Liquidity by Binkowski and L (2018). It is shown that if you add traded volumes it will improve the modeling of volatility (especially for US stocks) and you can gain a little more if you add the 2 other variables (see the figure).

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.