Bai-Perron test for structural change

[alvarezcc]

I wonder if someone could provide me an example of the application of these test. I checked in the manual and the user guide but I find nothing. The main doubt is about the list of regressors that should be provided to perform the test. It woul also be very helpfull to know if there is some kind of considerations to take account for. Thanks in advance for your attention!!

Christian

[TomDoan]

You just run it on the regression of interest. Your main concern is that the critical values aren't valid for non-stationary data. The calculation is correct regardless (in terms of finding the optimal break points), but the critical values aren't in the presence of unit roots.

[alvarezcc]

Is there a way to know which of the break dates is significant or better? Depending on the test specification one could get many break dates. In my case, I don´t have enough criteria to select the number of breaks other than that sugested by watching the graph. So how to choose the best break dates?

Christian

[TomDoan]

As I mentioned in a different thread, the point of these multiple break point tests, whether Bai-Perron or Zivot-Andrews or Lee-Strazicich, or ..., is to check a specification of a fixed coefficient model. If the change point analysis shows a break in the specification, it is highly unlikely that you would respond to that information by replacing your original model with the same specification estimated in two subsamples. Instead, you would look for a change in specification, maybe finding a different proxy, or transformation of a variable, or addition of some regressor which would somehow model the break.

[alexecon]

[...] Your main concern is that the critical values aren't valid for non-stationary data. The calculation is correct regardless (in terms of finding the optimal break points), but the critical values aren't in the presence of unit roots.

I would be interested to read more about why this is the case. Bai and Perron do not seem to discuss stationarity in their suite of papers. In fact, a simple ADF unit root test for the U.S. real rate they use in their 2003 paper does not reject the null (of a unit root).

Sorry to unearth a long-forgotten topic!

Many thanks for any thoughts.

Alex

[TomDoan]

Different co-author. See

https://estima.com/forum/viewtopic.php?p=2460#p2460

[alexecon]

Different co-author. See

https://estima.com/forum/viewtopic.php?p=2460#p2460

I understand that, in the presence of cointegration, the critical values herein have to be used for the double-max and sequential tests. Would I be correct in thinking that, in order to determine whether series are I(0) or I(1), a common ADF test would suffice (since Bai-Perron's test is for possible break(s) in the relationship and not the series themselves)?

If this is indeed the case and the ADF is fine, does it also apply to the simplest of permutations with just a (potentially breaking) intercept in the RHS? I.e. could we test error stationarity with it?

Thanks for any insights in advance.

Alex

[TomDoan]

There's an entire literature on testing for unit roots allowing for structural breaks. If you look at the unit root tests wizard, it includes three such tests (Zivot-Andrews, Lee-Strazicich, Lumsdaine-Papell).