Re: Letter form Bob Solow

["Lonnie K. Stevans" <acslks@xxxxxxxxxxxxxxxxx>]

I obviously cannot speak for Granger, but the procedure is to test for presence/absence of a
unit root before testing for cointegration.  It would seem to me that the process of just
arbitrarily taking nth-differences without testing for multiple unit roots is a
"meaningless exercise."

Lonnie K. Stevans
acslks

>>> pdavidso <pdavidso@xxxxxxx> 11/11/02 12:10PM >>>
Lonie Stevens wrote the following in response to my posting of the Bob Solow
letter on my article on nonoergodic uncertainty

This obviously transpired before the application of Granger's seminal work on
cointegration.


My response is:

Granger's cointegration work is irrelevant -- for it also assumes that if the
level is nonergorid, then one of the differences will turn out to be an
ergodic process.


Now I went thorough the fallacy of the cointegration argument on the pktnet
about 5 years ago.

The best sdource of a response is a book by a Russian Mathematican, A. Yaglom,
on the theory of stochasitc processes.

In the book (cited in my articles) Yaglom took a randon number generator and
generated two columns of random numbers.  Not surprisingly he foud a unit root
for these columns.  Yaglom, however,  also found that if he took the
nth-difference, he could ALWAYS get an apparence of statistical significance
(cointergration) between the nth-differences of the  two columns. [usually he
did not have to go beyong a 2nd or third difference)

His conclusion that tsking mindless difference to demonstate what Granger
calls cointergration is a meaninglesws exercise.

how would you or Granger explain this?


Paul

Paul Davidson
Editor, Journal of Post Keynesian Economics
University of Tennessee
SMC 523
Knoxville, Tennessee 37996-0550
phone # (865)974-4221; fax #(561)737-8262;
email pdavidson@xxxxxxx
http://econ.bus.utk.edu/davidsonextra/Davidson.html