unit roots test in the presence of breaks

[MC128]

Hi Tom and everyone,

I would like to ask about the procedure for testing unit roots in the presence of two structural breaks. Apart from the @lsunit procedure, do Rats offer any other procedure? In particular, do Rats have the testing procedure suggested by Lumsdaine & Papell (1997) "Multiple Trend Breaks And The Unit-Root Hypothesis,"?

Many thanks.

MC

[condor]

Here you can find a procedure written by Tom Doan.

[MC128]

Hi condor and everyone,

Thank you so much! I have compared the results given by the Lee-Strazicich and Lumsdaine-Papell test for two breaks. It seems that the result, particularly with regard to the break date, could be quite different....and even compared to the break date given by the one break test (i.e. lee-Strazicich and Zivot-Andrews). Is this normal?

And may I ask whether there is any rule governing the choice of lag length for the test? (the maxlag option if the option ttest is chosen). Most papers set a value of 8.....is there any rationale behind?

Thanks.

MC

[TomDoan]

Hi condor and everyone,

Thank you so much! I have compared the results given by the Lee-Strazicich and Lumsdaine-Papell test for two breaks. It seems that the result, particularly with regard to the break date, could be quite different....and even compared to the break date given by the one break test (i.e. lee-Strazicich and Zivot-Andrews). Is this normal?

Yes. Again, a reminder that these are really specification tests for unit roots without breaks. If there aren't any really sharp breaks, the position of the largest single break isn't going to be sharply estimated and it won't necessarily match the most extreme double break.

And may I ask whether there is any rule governing the choice of lag length for the test? (the maxlag option if the option ttest is chosen). Most papers set a value of 8.....is there any rationale behind?

Inertia.

Seriously, though, it's probably large enough to cover the bulk of cases.

[Anna]

Hey Tom and everyone,

I would like to ask a question as well and hope you are patient enough to answer.

I have done some structural break tests with some I(1) variables and found that most of them date the structural break at the very first or second observation, although a graphical analysis casts doubt on this.

First question: Is there any way to explain this?
Second question: Since Perron it is know that in order to determine whether a series is I(0) or I(1), we need to test for structrual breaks. But I wonder about the impact of the order of integration on the structural break tests like those mentioned above. Are they valid?

Comments are highly appreciated!

Thanks in advance!
Anna

[TomDoan]

There's usually a control parameter which prevents selection of a break too close to either end of the data set, or, in the case of multiple breaks, having the breaks spaced too closely. Otherwise it's very easy for a single outlier to be identified as a "break".

The Bai-Perron calculation is a mechanical procedure which works correctly whether the data are I(1) or I(0). Their critical values, though, are for I(0) data only.

[Anna]

Thank you very much Tom! You are very helpful. This is highly appreciated.

Could you spend a few words on the other structural break tests? What about the Andrews-Ploberger-test? In the AP-Breaktest I took the first lag of the dependent variable as regressor and my intuition says that this is basically ok. However, my intuition does not work very well when I am dealing with econometrics, so is this appropriate at all? Does it matter whether the data are I(0) or I(1)?

Another question: very recently, Perron and Kim (2009, Unit root tests allowing for a break in the trend function at an unknown time
under both the null and alternative hypotheses, Journal of Econometrics 148 1-13), have developed another unit root test which seems to be quite useful (see title which is self-explanatory). Is any code available for this test-procedure?

Thank you very much indeed, Tom!

Best wishes
Anna