Skip to content
All library documents

Choosing the Sampling Interval for Realized Variance GARCH

Article Quant Q&A · Author: s5s
Read the original

Summary

The document asks whether hourly or multi-hour realized variance can be used in a realized variance GARCH model. The author has minute-bar foreign exchange data and notes that the usual setup pairs daily returns with a daily realized volatility measure. With intraday observations available, the question is whether variance aggregated within shorter intervals is suitable instead.

No answer, model specification, empirical comparison, or fitting results are included. The useful issue is the relationship between the sampling frequency of realized variance and the time scale of returns and GARCH dynamics. The document does not establish that hourly variance can simply replace daily variance; such a change would need a model and data frequency aligned consistently. It is best read as an open modeling question about temporal aggregation, not as evidence for a particular interval or a validated estimation procedure.

Key ideas

  • Realized variance is commonly paired with daily returns in the model setup described.
  • Minute-bar data can support calculation of variance measures over shorter intervals.
  • The document asks whether hourly or multi-hour measures fit realized variance GARCH models.
  • It provides no answer or empirical evidence that a shorter interval is appropriate.

Tags

ForexVolatilityStatistics
Full text
# Realized Volatility + GARCH - can I use hourly realized volatility?


# Realized Volatility + GARCH - can I use hourly realized volatility?












I hav minute bar FX data and I am trying to fit a realized variance GARCH model using rugarch. This normally works by providing daily returns and daily realized volatility to the model.

Realized Variance is normally calculated as a daily statistic so you get a value for each day. Considering I am using minute bar data, I should have enough observations per hour to calculate hourly realized variance. Would hourly (or N-hourly) realized variance make sense with GARCH models?

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.