> From: bbol...@gmail.com
> Date: Fri, 1 Jul 2011 13:16:45 +0000
> Subject: Re: [R] How to fit ARMA model
>
> UnitRoot <akhussanov <at> gmail.com> writes:
>
> >
> > Hello,
> > I am having some problems with fitting an ARMA model to my time series data
> > (randomly generated numbers). The thing is I have tried many packages
> > [tseries, fseries, FitARMA etc.] and all of them giving very different
> > results. I would appreciate if someone could post here what the best package
> > is for my purpose.
> > Also, after having done the fitting, I would like to check for the model's
> > adequacy. How can I do this?
> > Thanks.
>
> It's hard to say without more detail -- we don't know what your
> purpose is (beyond the general one of fitting an ARMA model to data).
> I would say that when in doubt, if there is functionality in 'core' R --
> the base and recommended packages -- that it is likely to be the most
> stable and well tested. (Not always true -- there are some very good
> contributed packages, and sometimes the functions in R are missing
> some advanced features -- but a good rule of thumb.) So try ?arima.
>
> Another rule of thumb is that Venables and Ripley 2002 is a good
> starting point (although again not necessarily as extensive as specialized
> topics) for "how do I do xxx in R"? See Chapter 14.
>
I had to check my memory but IIRC wiki or other non-authoritative source
suggested that this is not much different than trying to design or analyze
an IIR (infinte impulse response) filter that is presumed excited with white
noise. In such a case, the
power spectrum could be interpretted in terms of zero and pole ( zeroes
of denomintator) of the transfer function giving you some graphical info
to help develop intuition.
It is not hard to make your own tests and sanity checks if you are just getting
started.
( hand typed as my 'dohs machine is busy running malware scans and this one R
install
has a few issues )
x<-runif(8192)
x1<-c(x[-1],x[1])
x2<-c(x1[-1],x1[1])
y=x+x1+x2
plot(log(abs(fft(y))))
makes bode plot ( if I have the name right) with zeroes at the root. Extension
to
IIR case should be obvious.
> As suggested in the help for ?arima, ?tsdiag is "a generic function
> to plot time series diagnostics"
>
> ______________________________________________
> R-help@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
