[R-sig-finance] Puzzled in arch estimation

[Spencer Graves]

	  Have you received a reply to this post?  I haven't seen one.

	  I ran the script you included with your post.  For the garch(1,1) 
model, I got an answer with a message "FALSE CONVERGENCE".  For the 
arch(1,0) model, I got a message, "Warning: singular information" with 
parameter estimates that raise questions in my mind about what the 
algorithm did.  You say the first one "Seems to be okay",  but the 
second one "breaks."  What do you mean by "breaks"?

	  I also did 'RSiteSearch("garch")'.  This revealed that the fSeries 
package also has a garch modeling function.  Have you looked at that? 
I'm sorry, but I've done very little with garch, and I don't have the 
time now to study it more deeply.

	  If you'd like more help from this listserve, PLEASE do read the 
posting guide! "www.R-project.org/posting-guide.html".  Anecdotal 
evidence suggests that posts more consistent with this guide tend to get 
more useful replies quicker.

	  hope this helps.
	

 > sessionInfo()
R version 2.2.0, 2005-10-06, i386-pc-mingw32

attached base packages:
[1] "methods"   "stats"     "graphics"  "grDevices" "utils"     "datasets"
[7] "base"

other attached packages:
  tseries      zoo quadprog
"0.10-0"  "1.0-3"  "1.4-7"
 >

Ajay Narottam Shah wrote:

> I have attached a time-series r. I find that I am able to use the
> garch() function in the tseries package to estimate a GARCH(1,1)
> model. But it fails to get convergence for the GARCH(1,0) -- i.e.,
> ARCH(1) -- model. I seem to think that the latter is actually a simple
> likelihood function and it's much easier to get convergence for it.
> 
> There's a mysterious statement towards the end of the time-series
> chapter of the MASS book which says that the garch() function requires
> the series to be mean 0 (a statement which I didn't find in the
> documentation of tseries). So I also did this one more way: To first
> estimate an AR(1) model, and focus on the residuals. Once again, I get
> the same pathology: GARCH(1,1) works but ARCH(1) does not.
> 
> I fed the same series to stata and it estimates AR(1)-ARCH(1) easily,
> so this doesn't seem to be a pathological case of bad
> data. Difficulties in convergence can, of course, always come about in
> realworld data. I wondered if I'm missing something fundamental.
> 
> Here's some code which illustrates my problem --
> 
> library(tseries)
> # take the definition of `r' ahead and then say
> 
> m.ar1 <- arima(r, order=c(1,0,0))
> m.garch11 <- garch(m.ar1$residual, order=c(1,1)) # Seems to be okay
> # BUT
> m.arch1 <- garch(m.ar1$residual, order=c(1,0))
> # breaks.
> 
> Alternatively, equally, one can do:
> 
> m.garch11 <- garch(r, order=c(1,1)) # Seems to be okay
> # BUT
> m.arch1 <- garch(r, order=c(1,0))
> # breaks.
> 
>     -ans.

> 

-- 
Spencer Graves, PhD
Senior Development Engineer
PDF Solutions, Inc.
333 West San Carlos Street Suite 700
San Jose, CA 95110, USA

spencer.graves at pdf.com
www.pdf.com <http://www.pdf.com>
Tel:  408-938-4420
Fax: 408-280-7915