Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: RE: xtunitroot ips gives blank results


From   Ben Gilbert <[email protected]>
To   [email protected]
Subject   Re: st: RE: xtunitroot ips gives blank results
Date   Sat, 25 Sep 2010 14:52:08 -0700 (PDT)

Hi Scott,
thank you very much for your help! For those of you who use xtunitroot, the outcome of this mystery might be helpful. The "Not available" result for t-bar and t-tilde-bar makes sense now that you point it out, but even when I "set trace on" and looked through the code I couldn't figure out why ALL the statistics were missing,

By running xtunitroot ips -depvar- on successively smaller collections of panels, I was able to isolate 3 panels (out of 10,000) that had constant values over all time periods. The sensor that was recording my data was broken in 3 locations and inputting the same value at every time step. A unit root test in this case regresses a vector of 0's on a vector of 0's.

For some reason, the fisher-type tests in xtunitroot can handle this issue when it averages test statistics over all the panels. With the IPS test, even one panel with constant values will cause the test to report blank statistics. Which is a good thing in my case, since I was able to catch some data that doesn't belong.

Thanks again,
Ben


Date: Wed, 22 Sep 2010 13:14:57 -0500
From: Scott Merryman <[email protected]>
Subject: Re: st: RE: xtunitroot ips gives blank results

A couple points.

First, since all the statistics are reported as missing try -set trace
on- (of course, there are no pvalues reported for t-bar and
t-tilde-bar, see example 4,  page 536 [xt] xtunitroot).  Second, the
reason the command reports "(Not available)" is that your data set is
unbalanced.  From page 536 [xt] xtunitroot,  paragraphs 2 and 3:

"First, we consider the serially uncorrelated case, which xtunitroot
assumes when you do not specify the lags() option.... For the case
where N is fixed, IPS used simulation to tabulate “exact” critical
values for the average of the ti statistics when the dataset is
balanced; these critical values are not available with unbalanced
datasets."

The "(Not available)"  result can be seen with:

webuse grunfeld,clear
xtunitroot ips invest
drop in 1
xtunitroot ips invest

Scott


On Wed, Sep 22, 2010 at 11:51 AM, Ben Gilbert <[email protected]> wrote:
> Thank you for your comments, Nick.
>
> I always look in the help file and manual first, then scour the stata-list archives on the WEB as well as other sources so as not to clutter the list with questions. In this case the manual and help files are not that helpful for this error.
>
> The command I typed was xtunitroot ips -depvar-
> After the command runs I get a table of blank results, formatted just as the xtunitroot output should look but with no test statistics and a Not Available message where the critical values usually display.
>
> I'm running Stata MP 11 for windows (64-bit), and as I said I have a large unbalanced panel (N=10,000, T ranges from 52 to 292) but no missing observations.
>
> Any suggestions would be greatly appreciated.
> Thanks in advance!
> Ben


      

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index