On Mon, Oct 7, 2013 at 5:39 PM, Riccardo (Jack) Lucchetti <r.lucchetti@univpm.it> wrote:

On Mon, 7 Oct 2013, Sven Schreiber wrote:

Am 07.10.2013 18:58, schrieb Allin Cottrell:

On Mon, 7 Oct 2013, Felipe L. Bhering wrote:

I've already seen it. The output of the irf function comes the confidence
interval. What , really , I want is not the interval of some alpha (some
confidence level) .

Is it possible to output the standard error used to make this confidence
interval?

No, because it doesn't use a standard error as such, it is based on the
"percentile method" (that is, the quantiles of the bootstrap iterations).
To get an estimated standard error you would have to do the bootstrap
yourself (though that might be added as an option at some point).

Just a follow-up on this: With the standard error alone you wouldn't be
able to construct any confidence intervals in this context, because in
general you cannot characterize the distribution in terms of just its
(first n) moments.

(But of course the estimated standard deviation of the simulated
distribution may still be a useful statistic. If such an option were to
be added, it should probably also return other moments such as skewness
and kurtosis.)

Follow-up on the follow-up: IRF confidence bands are a very tricky subject. There is quite a lot of literature on that. I suggest Helmut Lütkepohl's multiple time series book to give you a hint as to why standard errors may be quite a bad idea and provide you with a few literature pointers.

-------------------------------------------------------
  Riccardo (Jack) Lucchetti
  Dipartimento di Scienze Economiche e Sociali (DiSES)

  Universitą Politecnica delle Marche
  (formerly known as Universitą di Ancona)

  r.lucchetti@univpm.it
  http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users