Quandt Andrews Lr F or Wald F

[mscarlatos]

I am running Eviews Quandt-Andrews test (not using Newey-West standard errors) and get usual output on Lr F and Wald F. However, each statistic points to a different breakpoint as far apart as 60 quarters apart. I would like to know when it is appropriate to use the LR F or Wald F , ie., are there any rules of thumb to use?

[EViews Gareth]

Probably means that it is a close call between those two possible break points. If you save the LR and Wald statistics into your workfile and take a look at them, you'll probably see that they are quite flat, or at the very least, quite similar at those two points.

[EViews Gareth]

Of course I should add that if you're doing White standard errors, it is very possible that the two statistics will give different breaks, since the Wald takes the standard errors into account.

[mscarlatos]

If I am indeed using White standard errors and getting disparate breakpoints far apart in time, is there any rule of thumb in econometricians use in determining whether to go by LR F or Wald F? Some of my results have the LR pointing to an intuitively appealing breakpoint while other QA tests have the Wald signalling a correct date. I am tempted to use LR in certain cases and Wald in other..but this seems inconsistent. Must one stick to LR or to Wald throughout a research project?

[Mila]

Hi,
I'm facing a similar issue where the LR and Wald test statistics give different breakpoints with different p-values - very significant in the Wald case, when I use the White standard errors (there is mild heteroskedasticity in my equation).
I'm wondering though, if I use the White SEs, do the non-standard critical values in Eviews still apply? Or is there a tendency for over-rejection of the null.
My understanding is that the non-standard distributions derived by Andrews assumes fixed regressors and i.i.d. errors. And bootstrap values would need to be calculated otherwise.
I would really appreciate some clarification, at least as to the distributions in Eviews.
Many thanks!