## Bai and Perron str break test using Non-linear estimation

For technical questions regarding estimation of single equations, systems, VARs, Factor analysis and State Space Models in EViews. General econometric questions and advice should go in the Econometric Discussions forum.

Moderators: EViews Gareth, EViews Moderator

lakis
Posts: 13
Joined: Wed Oct 22, 2014 7:33 am

### Bai and Perron str break test using Non-linear estimation

Hi guys, following the Eviews suggestion on how to use the Bai and Perron test, it seems that initialy you estimate your model with LS and by listing your dependent and the regressors. How can I impose Non-linear least squares estimation before testing for breaks? Is that possible?

Many thanks

EViews Glenn
EViews Developer
Posts: 2636
Joined: Wed Oct 15, 2008 9:17 am

### Re: Bai and Perron str break test using Non-linear estimatio

It is not built in. The algorithms that one uses for the Bai-Perron test require linearity.

lakis
Posts: 13
Joined: Wed Oct 22, 2014 7:33 am

### Re: Bai and Perron str break test using Non-linear estimatio

The probles is that i get different estimates when i analyticaly write the estimated equation rather than just list the variables. So Is there any way we estimate using non linear LS and then conducting this test?? The problem is that Eviews does not allow you to conduct the test when you write the analytical specification of the estimated equation. It says that you have to list the variables.

EViews Glenn
EViews Developer
Posts: 2636
Joined: Wed Oct 15, 2008 9:17 am

### Re: Bai and Perron str break test using Non-linear estimatio

I don't understand the point about getting different estimates when you analytically write the estimated equation. Could you elaborate?

Oh, and on another note. If you only have a single break, the EViews implementation of the Quandt-Andrews tests allows you to test in nonlinear specifications. Note that the theory for performing this test on nonlinear specifications hasn't necessarily been worked out (though I suspect that it goes through).

lakis
Posts: 13
Joined: Wed Oct 22, 2014 7:33 am

### Re: Bai and Perron str break test using Non-linear estimatio

What my theoretical model requires is to estimate my model as: INTEREST_RATE_=(1-C(1))*(C(2)+C(3)*INFLATION+C(4)*OUTPUT_GAP)+C(1)*LAGGED_INTEREST_RATE
So If I write it like that in the 'Equation specification' i get different results compare to the case where I just list the variables in the 'Equation specification'.

lakis
Posts: 13
Joined: Wed Oct 22, 2014 7:33 am

### Re: Bai and Perron str break test using Non-linear estimatio

Please have a look at the the first sheet. I have 2 estimations..Please have a look at the 'Representation' of each estimation so you understand what I mean.
Attachments
multi_struct_break_tr_new.wf1

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 12462
Joined: Tue Sep 16, 2008 5:38 pm

### Re: Bai and Perron str break test using Non-linear estimatio

They're completely specifications. It is not surprising they result in different estimates.

lakis
Posts: 13
Joined: Wed Oct 22, 2014 7:33 am

### Re: Bai and Perron str break test using Non-linear estimatio

So the question now is how can I apply the Bai and Perron test in the INTEREST_RATE_=(1-C(1))*(C(2)+C(3)*INFLATION+C(4)*OUTPUT_GAP)+C(1)*LAGGED_INTEREST_RATE specification ????

EViews Glenn
EViews Developer
Posts: 2636
Joined: Wed Oct 15, 2008 9:17 am

### Re: Bai and Perron str break test using Non-linear estimatio

As I wrote earlier, Quandt-Andrews.

startz
Non-normality and collinearity are NOT problems!
Posts: 3551
Joined: Wed Sep 17, 2008 2:25 pm

### Re: Bai and Perron str break test using Non-linear estimatio

lakis wrote:So the question now is how can I apply the Bai and Perron test in the INTEREST_RATE_=(1-C(1))*(C(2)+C(3)*INFLATION+C(4)*OUTPUT_GAP)+C(1)*LAGGED_INTEREST_RATE specification ????

Why not just do the test listing the RHS variables? You should get the exact same regression.

lakis
Posts: 13
Joined: Wed Oct 22, 2014 7:33 am

### Re: Bai and Perron str break test using Non-linear estimatio

I need Bai Perron test only..Thats the problem..

lakis
Posts: 13
Joined: Wed Oct 22, 2014 7:33 am

### Re: Bai and Perron str break test using Non-linear estimatio

startz wrote:
lakis wrote:So the question now is how can I apply the Bai and Perron test in the INTEREST_RATE_=(1-C(1))*(C(2)+C(3)*INFLATION+C(4)*OUTPUT_GAP)+C(1)*LAGGED_INTEREST_RATE specification ????

Why not just do the test listing the RHS variables? You should get the exact same regression.

Because I am afraid that I would get different structural break results If I could write analytically the specification of the regression equation. Or not?

startz
Non-normality and collinearity are NOT problems!
Posts: 3551
Joined: Wed Sep 17, 2008 2:25 pm

### Re: Bai and Perron str break test using Non-linear estimatio

I think you get the same results, as the test depends on the sum of squared residuals. I'm not 100% sure.

EViews Glenn
EViews Developer
Posts: 2636
Joined: Wed Oct 15, 2008 9:17 am

### Re: Bai and Perron str break test using Non-linear estimatio

I just looked carefully at the spec, and Startz is right that's it's just a renormalized linear regression so all should be fine. The original poster saying that it was a nonlinear specification kind of threw a wrench into the discussion. The Bai-Perron tests are actually Wald tests, but since everything is linear, the results should be the same (as you will see if you perform a Quandt-Andrews and compare the results to the Bai-Perron with a single break).

lakis
Posts: 13
Joined: Wed Oct 22, 2014 7:33 am

### Re: Bai and Perron str break test using Non-linear estimatio

Guys...think I am getting crazy..Thats what I found in a published paper:

"Nonlinear least-squares estimates of the following equation: i=(1-ρ)(β+γ*p(t)+δ*y(t))+ρi(t-1)
where i is the Federal Funds Rate, p is forecasted inflation, and y is the output gap. "

First of all, the above model is linear or non-linear???? if its non-linear how do we estimate it in Eviews??

Much appreciated