[R-SIG-Finance] another garch question

[Spencer Graves]

      1.  Have you studied the 'garchFit' help page, especially the 
"Value" section? 

      2.  Are you familiar with the 'str' function? 

      3.  Between "1" and "2" and playing with simple examples, I'm 
usually able to get answers to most of my questions for things like 
this.  When that fails, I sometimes try simulated examples, e.g., 
generated via 'garchSim'.  Sometimes I work examples -- or parts of 
examples in something like MS Excel. 

      4.  The ultimately documentation is the source code.  You can get 
source code with comments from 'www.r-project.org' -> CRAN -> (select a 
mirror) -> Packages -> fGarch -> "Downloads:  Package source:  
fGarch_260.72.tar.gz".  Saying 'debug(garchFit)', then executing a call 
to garchFit will allow you to walk through the code line by line. 

      You might like to have documentation that is more user friendly, 
but you will NOWHERE find documentation that is as complete as the 
source code -- nor as easy to modify for slightly different purposes. 

      This may not answer your question, but I've given you here at 
least 4 different fish hooks.  You supply the bait, and you should be 
able to catch your fish. 

      Hope this helps. 
      Spencer

babel at centrum.sk wrote:
> Hello scientists
> How can I quantify the garch model? The equations are for example from book Analysis of financial time series by Tsay
> rt = 0.0067 + at,     at = σt * et 
>  σ2t = 0.000119 + 0.8059σ 2t−1 + 0.1941a2t−1
>
> fit = garchFit(~garch(1, 1), data =ret)
>
> What slot holds the et variable in R? (fit at residuals ??? I doubt) In every book I read there was only coefficients estimated, but no graphical output of modelled serie produced. So how can I said that for example model garch(1,1) is better than arch(2)? ArchTest,LM test and BDS test of the standardized residuals are the only options? or AIC, SIC?
>
> I tried 2 models
> fit1 = garchFit(~arma(1, 0)+garch(1,1), data =ret)
> fit2 = garchFit(~arma(1, 0)+garch(2,1), data =ret)
> and the fit1, fit2 at fitted were slightly different, so the volatility (garch term) must influence somehow to mean estimation. 
> Why so many books ended with coefficients estimation?And what tells me the predicted standard deviation about the future behaviour of return series? Are there good only for making confidence intarvals +-2*sd or to print simga.t and said so, this is my volatility?
> So how correctly use garch in modelling time series and comparing it with different methods if garch is for volatility and not for mean?
>
> Sorry for slow thinking :)))
>
> _______________________________________________
> R-SIG-Finance at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
> -- Subscriber-posting only. 
> -- If you want to post, subscribe first.