The RATS Software Forum

Dear Tom Doan,
I am testing a few macroeconomic time series for multiple structural breaks. Your Codes on RATS/Estima fourm are brillant and I was able to run them but encountered problem due to critical values. Multiple structural break results(Lumsdaine and Papell, 1997) have accepted the Unit root null hypothesis hence reversed my results. In order to make sure that I have no other problem, I have estimated the single structural break models for Brazil. Unfortunately, the results shows non-rejection of null hypothesis with the RATS programming while eviews confirms the presence of one structural break (rejecting unit root null hypothesis) either exogenous or endogenous breaks. I have some confident that this problem has arised because of the critical values rather than estimated coefficients. I have attached one of the series for you to view and check where I have made a mistake.(Even if RATS rejects the presence of two breaks, why it rejects the presence of one structural break? confirmed by the Eviews using exogeneous as well as endogenous structural breaks? I shall be thankful to you for your prompt reply,

Kind regards
Imran

First of all, that's a rather short data set to be thinking about allowing two structural breaks. More important, however, you just provided the data set with no program. How can I know what tests you ran? A couple of quick tries seemed to indicate that the unit root hypothesis is accepted with zero, one or two breaks, and quite easily, so I don't know where you are having a problem. However, your description seems to show that you don't understand the point of a Zivot-Andrews or LP unit root test---you're not testing for breaks, you're testing for unit roots allowing for breaks. It's possible for the breaks to be insignificant but still be enough to overturn the acceptance of a unit root, and it's also possible for the breaks to be significant and not overturn the result.

I would suggest that you read section 11.6 of the RATS (v8) User's Guide.

Thanks a lot for your prompt reply. Indeed, I am testing for unit roots allowing for structural breaks. So, unit root null should be rejected in every instance to accept the presence of broken trend stationary alternative. Since the null of unit root is not accepted even in one break case with my RATS program (attached), it raises doubts about the presence of single structural break.However, in case of two structural break your interpretations as well as literature supports the reversal of results in a high percentage including LP(1997).Your question provide insights of the problem but I am finding it difficult to understand exactly what do you mean by this " It's possible for the breaks to be insignificant but still be enough to overturn the acceptance of a unit root, and it's also possible for the breaks to be significant and not overturn the result"
I have attached the program for you to view.
thanks again,

Remain blessed,
Imran

Imran7692 wrote:Thanks a lot for your prompt reply. Indeed, I am testing for unit roots allowing for structural breaks. So, unit root null should be rejected in every instance to accept the presence of broken trend stationary alternative.

That's incorrect. The point of Perron's original paper, and the entire subsequent literature, is that stationary data with a small number of structural breaks "looks" like it has a unit root. So if you do a D-F test and accept the unit root, it might be because there are structural breaks instead. The L-P or Z-A or L-S or ... tests are really specification tests on the unit root testing procedure, to see if the unit root is rejected if you allow for structural breaks. It's not to test whether the data should be represented as trend-stationary with breaks---that interpretation has the null and alternative confused. Instead, it's to test whether the unit root (possibly with breaks) is wrong.

Imran7692 wrote:Since the null of unit root is not accepted even in one break case with my RATS program (attached), it raises doubts about the presence of single structural break.However, in case of two structural break your interpretations as well as literature supports the reversal of results in a high percentage including LP(1997).Your question provide insights of the problem but I am finding it difficult to understand exactly what do you mean by this " It's possible for the breaks to be insignificant but still be enough to overturn the acceptance of a unit root, and it's also possible for the breaks to be significant and not overturn the result"

You're not testing for the presence of a single structural break. You're testing the unit root hypothesis. With your data set, you accept a unit root (easily) allowing for 0 breaks, 1 breaks or 2 breaks. What you should conclude is that the data has a unit root.