More Late Night Rambling, Sponsored by GARCH

3 Aug

Fun experimenting with GARCH modelling using the rgarch package in R on US Tsy yield returns. Was originally a late night excursion to test the ability of using the multicore package to parallelize GARCH fits, but turned into much, much more.

Interesting thing to note and it’s not fully captured is the “risk explosion” seen in the earlier parts of the returns with respect to risk/volatility forecasts: that was the tail end of the 2008 crisis.

Difference in the GARCH spec:

Top: Mean Model, using standard normal distribution, with ARMA(1,1). Bottom: Same Model, ARMA(0,0)


3 Responses to “More Late Night Rambling, Sponsored by GARCH”

  1. professorpinch August 4, 2010 at 12:39 am #

    Interesting that leads & lags are significant and can be either positive or negative. Seems to make specifying a model a challenge.

    Whenever you can talk about it more, I’m all ears.

    Nice work.

  2. chibondking August 4, 2010 at 1:30 pm #

    Thanks for the comment. Yes, model specification is indeed a tricky area.

    It’s difficult at least in my experience (and I’m rusty on GARCH related stuff, having last looked at it in 2007/08) to fit one model to a particular asset class. I made the cheap shortcut (lazy) of assuming a GARCH(1,1) and ARMA(0,0) for all 4 series, even though they have different characteristics. The Cross correlation chart is a bit misleading, I should have put the ACF plot of the observations, which shows a bit more information (regarding possible autocorrelations of the underlying series, etc.). While flipping thru the charts, I have noticed some interesting things that require more digging into.

    But yes, picking a model is … not easy in my limited experience.. especially if you’re going to try to replicate the model across a lot (in my mind more than 10) of underlyings. But, some of the more experienced stats/quants out there can surely chime in.

    • professorpinch August 4, 2010 at 8:27 pm #

      Well, GARCH(1,1) seems to work so much I wouldn’t call it lazy as much as I’d call it expedient 😉

      Correct me if I’m wrong, but don’t a lot of ARMA models also use a (1,1) spec? It’s been over 5 yrs since I really looked at any time series texts or papers, so no matter how rusty you are on this, I’m rustier.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: