Using Intraday Volatility Estimators to Predict Daily Returns

Summary

The document asks whether volatility measures calculated from intraday price data can help predict close-to-close squared returns. The intended use is estimating the profit and loss of a gamma scalping strategy that trades once daily, with trading assumed to occur at the close. It names Rogers–Satchell, Garman–Klass, and Yang–Zhang as examples of intraday estimators.

The author reports that an exponentially weighted moving average of historical Garman–Klass volatility predicts the next day’s Garman–Klass volatility well, but finds that intraday-based estimates do not similarly predict next-day close-to-close squared returns. The document raises the question of whether these measures can be adapted for that target, but provides no answer, empirical details, or proposed method. The central caveat is that forecasting one volatility measure does not establish usefulness for a different return target.

Key ideas

• The application described is forecasting daily squared returns for a once-daily gamma scalping strategy.
• Rogers–Satchell, Garman–Klass, and Yang–Zhang are cited as estimators based on intraday data.
• An exponentially weighted average of historical Garman–Klass volatility reportedly predicts its next-day counterpart well.
• The author reports no comparable success when using intraday estimates to predict close-to-close squared returns.
• The document poses the adaptation question but does not resolve it.

Tags

# Are intraday volatility estimators useful for close-to-close predictions


# Are intraday volatility estimators useful for close-to-close predictions












I am interested in predicting the PnL of a gamma scalping strategy which trades only once per day. For simplicity, let's say we can always trade at the daily close. So, what I need to predict are the daily squared returns. I saw that there are various estimators based on intraday data, for example Rogers Satchell, Garman Klass, Yang Zhang, etc.

I can see that these estimators are useful when I am actually trading intraday. Also, for example if I do an EWMA based on historical Garman Klass volatility, I get very good results in predicting the next day Garman Klass volatility. However, this is not the case if I use an estimator based on historical intraday data to predict next day close to close squared returns.

Are intraday estimators useful at all for my application and is there a way to make them work to predict close to close?

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.