Negative chow test statistic

[john anania]

I generate a chow break point test for a system of equations and keep getting some negative values for my test statistic. Can you tell if there is anything wrong with my code:

'*********************** Systems of equations estimation and testing*********************************************

system hm1
hm1.append lhpr = c(1)*lypdk +c(2)*lreal_stock +c(3)*rm5yr
hm1.append lresk = c(4)+c(5)*lhpr +c(6)* rprime+c(7)*lwager
hm1.append inst lypdk rprime rm5yr lwager
smpl 1982q1 2011q4
hm1.fiml
scalar ssr1= hm1.@ssr(1)
scalar ssr2= hm1.@ssr(2)

'Rolling chow break point test for demand cointegration vector

for !horizon=1 to 79
smpl 1982q1 1985q1+!horizon
hm1.fiml
scalar ssr11a= hm1.@ssr(1)
scalar ssr22a= hm1.@ssr(2)
smpl 1985q1+!horizon+1 2011q4
hm1.fiml
scalar ssr11b= hm1.@ssr(1)
scalar ssr22b= hm1.@ssr(2)
smpl 1985q1+!horizon 1985q1+!horizon
genr chow1 = ((ssr1-ssr11a-ssr11b)/3)/((ssr11a+ssr11b)/(120-6))
genr chow2 = ((ssr2-ssr22a-ssr22b)/4)/((ssr22a+ssr22b)/(120-8))
next
stop

[EViews Glenn]

I'm pretty sure that the system objective function estimated with FIML doesn't guarantee positivity of the statistics that you've produced. I don't believe they are valid for the hypothesis you wish to test.

[john anania]

How would you test for structural breaks in such a system?

[EViews Glenn]

The simplest would be an LR test.

[john anania]

I'm a novice. Could you elaborate. thanks

[EViews Glenn]

http://en.wikipedia.org/wiki/Likelihood-ratio_test

The restricted likelihood comes from the full period specification.
The unrestricted likelihood is the sum of the likelihoods for the two partial period specifications.

Perhaps other will chime in.

[john anania]

Thanks. If you have the time, could you give me the code to calculate the test statistic.