[Gretl-users] ADF test

[Allin Cottrell]

On Thu, 16 Nov 2006, john w wrote:

> I tried to do ADF tests with and without Test down option. Using 
> always the same variable.

> Here are the results...

(And the results differ for lag-order 0, depending on whether the 
test-down option is used or the 0 value is entered as a fixed 
parameter.)

I've found the bug: in the case of automatic testing down, if we 
got to the point of just one lag remaining, and if this lag was 
not significant, it wasn't actually deleted from the test model as 
should be.  The bug was specific to the case where the lag order 
"wanted to go to zero"; in this case the result from manually 
setting the order to 0 was correct, and the result from automatic 
testing-down was wrong.

This is now fixed in CVS and I'll put out a Windows snapshot 
shortly.

Allin.