Skip to content
All library documents

Designing GARCH and HAR Volatility Forecasts with Continuous Sentiment Data

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

Summary

The document raises research design questions for comparing two sentiment datasets as predictors of volatility. It considers whether to include both GARCH and HAR models to broaden the analysis, and how to use sentiment measured continuously when the HAR model uses intraday observations only during the trading session. The prompt also asks whether exponential decay could summarize sentiment arriving outside market hours and on weekends.

No answers, model specification, results, or evidence are provided, so the document does not establish that using both models or applying exponential decay is appropriate. The practical issue is aligning sentiment information with the timing and sampling frequency of volatility observations. Researchers would need to define the mapping from off-hours sentiment to trading-period measures and test alternatives, while keeping the limits of each forecasting setup clear.

Key ideas

  • The proposed study compares two sentiment datasets as predictors of volatility.
  • The research question considers using both GARCH and HAR specifications.
  • The HAR inputs are intraday observations limited to the trading session, while sentiment is available around the clock.
  • Exponential decay is posed as a possible treatment for sentiment outside market hours, but no answer or empirical evidence is given.

Tags

VolatilityMachine learningStatistics
Full text
# Volatility forecast, GARCH and/or HAR?


# Volatility forecast, GARCH and/or HAR?












I am comparing 2 different sentiment datasets in predicting volatility, I would have 2 questions regarding this.

1, Does it make sense to use both GARCH and HAR models for this research to cover a wider scope?

2, For the HAR model I use intraday-minute level- data, for minutes that fall in the trading day, so 9:30-16:00. My sentiment values are 24/7.

How can I handle sentiment values outside trading hours and weekends in the model? Would an exponential decay make sense for these timeframes?

Appreciate all the replies!

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.