st: how zandrews command chooses k lags by t-stat test

[Kit Baum <baum@bc.edu>]

Mark said

I have been reading thru the code for the zandrews command which does an
ADF style unit root test with one break.  I am wondering whether one
detail is actually implemented as was described in the paper.

In Zivot and Andrews (1992), p. 255, it says that "It is important to
note that the number of extra regressors, k, required for the ADF
regressions was allowed to vary for each tentative choice of lambda."

The way I read this, you need to re-calculate k for each tentative break
date.  Computationally intensive, but makes sense.  But as I read the
code for the zandrews command, it appears to calculate k only once at
the beginning, in fact without any breaks at all.   Then it seems to use
that same value for every regression over which you minimize the
t-statistic of the lagged regressand, which you test to identify a unit
root.

I am wondering if I am interpreting this correctly.  If so, it seems
like a nice heuristic, but different from what is in the original  
paper....


Mark is quite right. As the help file for -zandrews- indicates, this  
routine was translated from RATS code (available from http:// 
www.estima.com/procs_alpha.shtml) more or less wholesale, with one or  
two corrections. That RATS code does not incorporate the recommended  
determination of lag length as separate for each \lambda (break  
point). Although the original JBES article was consulted in the  
development of -zandrews-, the excellent point Mark raises was not  
incorporated into the Stata code. I will put that fix on the queue of  
desirable enhancements to my routines. Thanks to Mark for bringing it  
to my attention.


Kit Baum, Boston College Economics
http://ideas.repec.org/e/pba1.html


*
*   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/