Prais-Winsten Procedure

[Rod]

I am having trouble getting a clear idea of how to calculate rho for the P-W iterative procedure. Kmenta (Second Edition, page 319 ) says
rho=sum(2,n) e(t)*e(t-1)/ sum(3,n) e(t-1)**2 while virtually every other source I can find on the internet says regress e on e lagged 1. Obviously there is a significant difference between the two approaches. [It does appear that most software is using a 1 period lag to get rho.] <= this was not properly stated. I meant that most sum from 2 and not 3. Of course, everyone uses a 1 period lag!

[EViews Glenn]

Is that really a 3 in the denominator? And if it's a 2, why is there a significant difference?

[Rod]

Yes, it is really a 3, which is why I have the question. I also should have said that most appear to sum from 2, not 3

[EViews Glenn]

I think it's a typo.

[Rod]

Actually, I think it is a mistake because the only way I can reproduce the results is by using the printed formula!
rho=sum(2,n) e(t)*e(t-1)/ sum(3,n) e(t-1)**2

[EViews Glenn]

That last statement is ambiguous...

[Rod]

rho=sum(2,n) e(t)*e(t-1)/ sum(3,n) e(t-1)**2

No matter, I have moved on to looking at other sources.

[EViews Glenn]

That's fine. Just to clarify, I wasn't certain what "results" you were trying to reproduce.