Bai Perron testing

[antipa]

dear all,

i have eviews8 and am wondering whether and how i can construct confidence intervals around the break dates? also, i am not able to run the test procedure for an ar(1) process.
thank you very much in advance

[EViews Glenn]

We currently don't offer the computations for the breakpoint confidence intervals. They are not particularly hard to do (there are discussions in the references we provide on how to do this), but we were not impressed with the behavior of the existing estimators in our time experimenting with them so we decided to hold off for the moment. We may revisit the issue if there is demand.

As to AR models, the Bai-Perron computations requires that we estimate roughly T^2 different regressions, which can do easily for linear models but not for nonlinear. The problem becomes even more acute if you have partial break models as each of the T^m models must be estimated separately in that case. This is not practical using existing algorithms given the current state of computing power.

[antipa]

Thank you very much for your quick answer. could you please tell me more precisely where to find the discussion on the confidence intervals?

Thanks a lot again!

[EViews Glenn]

I don't have the references at my fingertips, but the Bai-Perron (Econometrica and Journal of Applied Econometrics) articles we cite in the manual are the first place I would look.