Chow/Quandt-Andrews breakpoint test for VARs

tbui · Postby **tbui** » Tue Mar 15, 2011 6:58 pm

Hi,

I understand that those tests are for equation only. If I'm interested in testing structural break in a VAR, equation by equation, what'd be the best way to do it? I'm thinking about breaking up a certain VAR into n single equations. What's the best command (makemodel or makesystem)? I'm having trouble getting single equation out of either a system or a model to apply those test to. Any suggestions?
Thanks,

Trung

EViews Gareth · Postby **EViews Gareth** » Tue Mar 15, 2011 7:50 pm

You could use makesystem to turn your VAR into a system, add some dummy variables, then do a Wald test

tbui · Postby **tbui** » Tue Mar 15, 2011 8:27 pm

How do you do a Wald test? I don't understand these steps. Let's say I have a bivariate VAR(1) of ffr and oilprice. I want to test if there is a break in each equation (1) ffr regression on lag oilprice and (2) oilprice regression on lag ffr. I want to put them into a program so that I can change/track them later.
Thanks,

T

EViews Gareth · Postby **EViews Gareth** » Tue Mar 15, 2011 8:50 pm

You have to make the system first, then estimate it, then do a wald test on the system.

tbui · Postby **tbui** » Wed Mar 16, 2011 7:08 am

I have tried that, getting system estimated. However, what I want is to perform the breakpoint tests Chow/Quandt-Andrews/Bai-Perron on the equations in the system. Is there anyway to handling individual equation in the system? I can run a loop and specify each equation from a VAR and test, but that isn't efficient.
T

EViews Gareth · Postby **EViews Gareth** » Wed Mar 16, 2011 7:56 am

When you turn the VAR into a system, you will have an equation, with separate coefficients, for each equation in the VAR. You can then specify the Wald test using those coefficients.

Bspencer11 · Postby **Bspencer11** » Mon Apr 04, 2011 12:27 pm

Gareth, I followed your instructions and was able to conduct a Quandt-Andrews test for an unknown break date for a single equation taken from a VAR. After doing some further work, however, I found that I needed to take first differences for a few of my series in order to deal with issues of non-stationarity... Attempting to perform the same procedure to test for a break date, (and using my newly revised data set), I was given the error message "Specification leads to zero or invalid breakpoints." This would be all well and good - it's entirely possible that no breakpoints exist within the data set - except that when I perform individual Chow tests for dates that I am especially interested in, I AM finding evidence of potential breaks. Any ideas as to what could be causing this error? I'm currently using Eviews 6 (Student version) and have attached my workfile below.

EViews Gareth · Postby **EViews Gareth** » Mon Apr 04, 2011 1:05 pm

You'll get that error message if any of the dates cause an estimation to fail, such as if singularities occur etc...

Bspencer11 · Postby **Bspencer11** » Mon Apr 04, 2011 2:13 pm

Hmm... all of my data points are defined, though I suppose that it's possible that the Chow test statistic is undefined for some date or another, and is leading to an error for the overall quandt/andrew's test. Is there any way to know which particular observation is causing the error? I guess I could run a Chow test for each observation, but it would be pretty time consuming...

Thanks so much for your help.

EViews Gareth · Postby **EViews Gareth** » Mon Apr 04, 2011 2:15 pm

Unfortunately not.

Tripeaky · Postby **Tripeaky** » Thu Aug 04, 2011 10:30 pm

I found out a lot of useful info. That fosterd internet marketing tips to make my problem fixed.:)

EViews.com