Bookmark and Share

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

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

st: Conservatism in analysis and modeling for AR(1) behavior

From   Thomas Jacobs <>
To   StataList <>
Subject   st: Conservatism in analysis and modeling for AR(1) behavior
Date   Sat, 19 Jun 2010 18:42:49 -0500

I am trying to perform an analysis of a variable at 5 different tenors
that exhibits borderline autocorrelation.  In panel, running xtserial
leads to rejection of the null of no serial correlation in the errors
at the 1% level of significance.  Yet in individual regressions by
tenor, I can only reject the null of no serial correlation in one of
the 5 tests at the 10% level using estat bgodfrey.  In performing an
integrated analysis I have the following choices.  (Note that I cannot
reject the panel level null of no fixed effects when running xtregar.)

At the panel level:
Run xtpcse with a single common source of autocorrelation within each
of the panels (option corr(ar1))
Run xtpcse with separate sources of autocorrelation within each panel
(option corr(psar1))

At each tenor:
Run prais with vce(robust) if I could get it to work!
Run regress with vce(robust)

Based on the test results, that at a given tenor it is difficult to
find serial correlation in the errors yet it appears to clearly exist
in a panel level test, what is the "conservative approach" in each of
the potential choices for reporting combined results :

assuming a separate source of serial correlation exists at each tenor
and run xtpcse, corr(psar1) along with prais for individual tenors


assume it only exists in panel across tenors and not in individual
tenors and run xtpcse, corr(ar1) and regress for individual tenors

Thanks for any help anyone might provide.


Thomas Jacobs
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index