Quandt Andrews Test Vs Chow Test

[madas]

hello there!

just a quick question!
i run the Quandt Andrews unknown breakpoint test and all three statistics fail to reject the Ho of No Breakpoints (at all significant levels). However, if i use the date when the max LR F stat appears to run a Chow test, the Ho of No breaks is rejected at 5%.
which one should i consider more reliable? and why is there a difference?

thanks in advance and well done for the great job! this forum is a treasure!

[Mila]

Hi Madas,
Have you figured out why is there a difference yet?
I'm also facing a similar problem. I wonder if has to do with the distribution for the test statistics. Plus, for Quandt Andrews, the trimming affects the result, whereas with Chow, all the observations are used.
I'm not sure if that's the answer though.

[EViews Glenn]

The distributions are different when you assume that the date is known versus when the date is unknown. The QA test is doing lots of Chow tests and picking one. That's not the same as doing a single Chow Test and doing lots of tests affects the distribution of the resulting staitstic. The references in the manual discuss this in detail.

[Mila]

Thanks very much Eviews Glenn!