Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Re: levinlin


From   Kit Baum <baum@bc.edu>
To   statalist@hsphsun2.harvard.edu
Subject   st: Re: levinlin
Date   Thu, 5 Aug 2004 10:49:44 -0400

On Aug 5, 2004, at 2:33 AM, Falko wrote:

Dear statalisters,
I have a question concerning the levinlin-ado file (panel unit root test,
ssc install levinlin).
The option 'nocons' omits the constant term from the individual Dickey-
Fuller regressions. Is it possible to test for the joint significance of
the constant terms? My problem is that I do not know how to address these
coefficients after the estimation, because they are not returned in the
ereturn list.
Whenever one estimates a unit root test, one must decide whether constants and/or trends are to be included in the specification of the Dickey-Fuller regression. A simple D-F model with constant (but without trend) would be

\Delta y_t = a + b y_t-1 + e which in the case of stationarity of the series implies a steady state of

y = a / (1+b) + e

If the mean of (level) y is demonstrably nonzero, then it is unreasonable to omit a constant term from the D-F regression, since that would imply that under the alternative y would be meanzero (like e). Under the null hypothesis, b=0 and a is the drift term in the random walk. One could have a pure RW without drift, of course, but if you're looking to reject H_0, you need to have a H_1 that has a chance of fitting the DGP of the observed data.

So unless your _level_ variables are meanzero, leave the constant term in there, just as you would normally include a constant term in a static regression equation.

Kit
coauthor levinlin

*
* 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/




© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index