SERVING THE QUANTITATIVE FINANCE COMMUNITY

volatilityMan
Topic Author
Posts: 104
Joined: January 16th, 2015, 6:06 pm

Volatility modelling

I wish to do some in-sample volatility modelling on stock indices as well as individual stock.

Initially, I did a GARCH(1,1) on daily stock returns [rolling window]. This did not always come out well. The parameters were very unstable especially for those stocks which exhibited large fluctuations.

So my question is:

- If one wants to do "simple" in-sample volatility modelling, what approach would you suggest?
- Both on a 5-min. interval but also on daily data...?

Alan
Posts: 9741
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Re: Volatility modelling

My experience is that GARCH parameters are reasonably stable on broad-based indices, using lots of daily data (say 20 year stretches) and avoiding windows with the '87 crash.

On single-name stocks, just as a guess, the instability may be largely driven by the earnings day moves. In the U.S., earnings releases are almost always pre or post regular session. You might first try modeling just the open-to-close daily returns with GARCH (instead of close-to-close) and see if that cures the issue. Then, decide how you want to handle the close-to-open returns on earnings day -- some kind of seasonal dummy is my first reaction.

Re the 5-min data, if you go my suggested open-to-close route, you could always use the realized variance from the 5-min returns as an improvement over the standard $\epsilon^2_t$ term in std. daily GARCH(1,1)

volatilityMan
Topic Author
Posts: 104
Joined: January 16th, 2015, 6:06 pm

Re: Volatility modelling

I tried all of it and your suggestion worked well. In the end I decided just to go with the non-parametric realized variance solution. It works and does the job plus I avoid a lot of issues. When possible, the non-parametric solutions are my favourites.

Thanks again.