[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Re: seasonal adjustment

From   "Nick Cox" <[email protected]>
To   <[email protected]>
Subject   st: RE: Re: seasonal adjustment
Date   Tue, 12 Aug 2008 15:36:57 +0100

My statement was general in respect of various more-or-less complicated
and supposedly standard procedures. 

But naturally I am grateful for the specific facts Kit adduces in the
case of X12. 

[email protected] 

Kit Baum

One statement below is incorrect. The source code for US Bureau of the  
Census procedure X12 (be it ever so arbitrary) is publicly accessible:

in fact I raised that issue with someone at StataCorp some time ago,  
pointing out that Bill Gould had helpfully written a primer on how to  
convert Fortran to Mata in the SJ, so why didn't someone take this on?  
The only issue is that the source code for X12 involves tens of  
thousands of lines of Fortran, and perhaps understandably that has not  
been a high priority at StataCorp.

It would be a great service to the user community if someone did that,  
though, as here in the US timeseries massaged by X12 are ubiquitous.  
In order to show the potential weaknesses of that procedure, one must  
be able to replicate it, and it would be very attractive to be able to  
do so in Stata without resorting to external programs.

On Aug 12, 2008, at 02:33 , Nick wrote:

> On your main question, I think the answer is No. You can -findit
> seasonal- to see what is available and nothing like X12 is evident to
> me.
> I don't doubt that there are Stata users who would like to apply one  
> or
> more of these seasonal adjustment procedures in Stata, but they appear
> highly unStataish to me, and it seems no surprise that there are no
> Stata implementations.
> These procedures seem to be very complicated and relatively  
> inflexible.
> They tend to wire in arbitrary empiricisms and be very much of their
> time. The idea that there is even a family of adjustment procedures  
> that
> will clean data regardless of (the rest of the) data generating  
> process
> seems not to march well with current time series ideas. Whenever I  
> poke
> around I tend to get the impression that only the source code would be
> an adequate source for precisely what is being done, and that may  
> not be
> easily accessible. For these and yet other reasons I am not surprised
> that no user-programmers have implemented any such.

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index