[R-SIG-Finance] Question about garchSim and garch

[Patrick Burns]

tom soyer wrote:

> Patrick, I am sorry maybe I didn't explain it well. I was thinking 
> using arma to estimate the mean, and garch for the conditional 
> variance. Does that make sense?


Yes, that makes sense, and that is what I was talking about:
it seems to be the case that estimating the mean model and
the conditional variance model separately tends to give you
a similar answer as estimating them both in a single procedure.

Pat

>  
> With regard to comparing models, do you, or anyone else know how to 
> build news impact curves in R?
>  
> Thanks!
>
>  
> On 2/3/08, *Patrick Burns* <patrick at burns-stat.com 
> <mailto:patrick at burns-stat.com>> wrote:
>
>     tom soyer wrote:
>
>     >Thnaks Spencer. I am glad I am not the only one that find garch
>     strange. I
>     >guess I will give up on it too. It seems that garchFit and
>     garchSim are very
>     >good. They have been giving me good results so far.
>     >
>     >Thanks for the tip on how to specify arma + garch model. I found
>     this paper
>     >also very
>     helpful:http://www.itp.phys.ethz.ch/econophysics/R/pdf/garch.pdf.
>     >
>     >Do you know how to specify arma + egarch model in R? Is it even
>     possible in
>     >R without installing Ox?
>     >
>     >
>
>     In my experience ARMA estimation and garch estimation are
>     suitably robust to each other.  It is definitely second prize to have
>     to estimate one and then the other, but your results are unlikely to
>     be all that different than if you did it "right".  (I'd love to
>     hear of
>     any counter-examples.)
>
>     Patrick Burns
>     patrick at burns-stat.com <mailto:patrick at burns-stat.com>
>     +44 (0)20 8525 0696
>     http://www.burns-stat.com
>     (home of S Poetry and "A Guide for the Unwilling S User")
>
>     >
>     >On 2/2/08, Spencer Graves <spencer.graves at pdf.com
>     <mailto:spencer.graves at pdf.com>> wrote:
>     >
>     >
>     >>Hi, Tom:
>     >>
>     >>     The file 'scripts\ch03.R' in the FinTS package includes a brief
>     >>description of attempts to use garch{tseries} and
>     garchFit{fGarch}.  I
>     >>don't understand either function very well, but I got answers from
>     >>'garchFit' that seemed to match some of the published results in
>     Tsay;
>     >>I gave up on 'garch'.
>     >>
>     >>     Since 'garchSim' and 'garchFit' are both in 'fGarch', I would
>     >>expect that it should be moderately easy to simulate something,
>     plot the
>     >>result, and see for yourself.  Chapter 3 of Tsay (2005) gives a
>     >>reasonable overview of GARCH and related models with several
>     examples.
>     >>The companion script\ch03.R is far from complete but might help.
>     >>
>     >>     You may find the following example from 'ch03.R' of interest:
>     >>
>     >>library(FinTS)
>     >>data(sp500)
>     >>library(fGarch)
>     >>spFit30.11 <- garchFit(sp500~arma(3,0)+garch(1,1), data=sp500)
>     >>
>     >>     This specifies an arma(3,0) mean model with garch(1,1) noise.
>     >>This syntax is buried in the 'garchFit' help page.
>     >>
>     >>     Hope this helps.
>     >>     Spencer
>     >>
>     >>tom soyer wrote:
>     >>
>     >>
>     >>>Hi,
>     >>>
>     >>>I am new to GARCH and I am trying to figure out how to use R's
>     garchSim
>     >>>
>     >>>
>     >>and
>     >>
>     >>
>     >>>garch, and I am a bit confused. I am hopeing that R finance
>     experts can
>     >>>
>     >>>
>     >>help
>     >>
>     >>
>     >>>me understand them better. If we look at the definition of
>     GARCH(1,1),
>     >>>there should be two equations:
>     >>>[1]: Y(t) = c + e(t), and
>     >>>[2]: sigma^2(t) = a0 + a1*e^2(t-1) + b1*sigma^2(t-1)
>     >>>
>     >>>So, I would expect any garch simulation function to four
>     parameters: c,
>     >>>
>     >>>
>     >>a0,
>     >>
>     >>
>     >>>a1, and b1. But take a look at the garchSim, it has only three
>     >>>
>     >>>
>     >>parameters:
>     >>
>     >>
>     >>>model = list(omega = 1.0e-6, alpha = 0.1, beta = 0.8). I assume
>     here
>     >>>
>     >>>
>     >>that
>     >>
>     >>
>     >>>omega = a0 in [2], alpha=a1, and beta=b1. If so, then it seems
>     that in
>     >>>garchSim, c, the constant (or the mean) in [1], is always
>     assumed to be
>     >>>zero. Does anyone know if this is true? I just want to make
>     sure that I
>     >>>understand exactly what I should expect from the output of the
>     garchSim
>     >>>function.
>     >>>
>     >>>Also, I have a similar question about garch. It seems that the
>     >>>
>     >>>
>     >>coefficients
>     >>
>     >>
>     >>>estimated by garch(x,order=c(1,1)) are a0, a1, and b1. Like
>     garchSim,
>     >>>
>     >>>
>     >>there
>     >>
>     >>
>     >>>is no c, the mean. So does this mean garch also assumes zero
>     mean and
>     >>>
>     >>>
>     >>thus
>     >>
>     >>
>     >>>actually fits model [2] instead of both [1] and [2]?
>     >>>
>     >>>Thanks!
>     >>>
>     >>>
>     >>>
>     >>>
>     >
>     >
>     >
>     >
>     >
>
>
>
>
> -- 
> Tom