# Re: st: RE: filters in stata

 From "Austin Nichols" <[email protected]> To [email protected] Subject Re: st: RE: filters in stata Date Tue, 4 Dec 2007 12:16:03 -0500

Juliusz Jabłecki <[email protected]>:

I disagree that replacing missings by zeros in an infinite sum should
be fine.  I think the programming is fairly easy, but given that the
task seems like a bad idea, not worth the time.  Consider what such a
method would do to a time series of 10 obs with b=0.5, picking some
arbitrary nonzero X_t values for t in [1,10].  The obs near t=1 and
t=10 will be smoothed toward zero, for no good reason.

You could get very close to what you claim to want by running -lpoly-
(possibly within panel using -statsby- perhaps) to do local mean
smoothing using a Gaussian kernel and an appropriately chosen
bandwidth, I think.  But again, it's not clear to me that this is a
good idea.

On 12/4/07, Juliusz Jabłecki <[email protected]> wrote:
> > The difficulty here is the infinite window width. In practice you
> > cannot set up an infinitely long dataset in Stata, so you
> > must compromise somewhere. Is there any scope for doing this
> > in the frequency domain? I never got as far as doing Fourier
> > transforms in my head.
>
> Sorry for not making that clear from the outset. Lucas's idea was to replace
> the missing observations by zeros, and that should be fine, except I don't
> have a clue as to how to actually program such a filter. Again, I'm trying
> to smooth a series X_t with the following filter (dependent on b):
>
> X_t(b)=[1-b/1+b]*[\sum(k=-inf/inf)b^|k|*X_(t+k)], 0<b<1
>
> and we replace the infinitely many missing observations by zeros.
> Any ideas?
> Juliusz

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/