Different output with rats 7 and 8

[Aktar]

Hi,

I have two version of rats: the 7 and the 8. When i perform MS-TVTP model with both,i do not have the same output! (inferred probabilities, transition probalbilities, coefficients...)it is the same prg but open with two different version of rats!

Maybe it is a mistake of myself, but the prg is the same.

I can attach my files if necessary

[moderator]

We'd probably need to see exactly what you are doing to have any idea why you are seeing differences.

Please email the code and data files to

support@estima.com

Be sure to include your full name, your RATS serial number, and the specific version you are using (do Help-About RATS to check the version number).

Regards,
Tom Maycock

[TomDoan]

Markov switching models are notorious for having multiple modes, and that's what you have. In this case, very slight differences in the way the simplex algorithm worked in 7.0 vs 8.0 causes it to end up on a different "hill". In this case, version 8 ends up with the higher likelihood, but there's no guarantee that that would always be the case. The difference in function values between the two modes is quite small (-254 vs -257) so if you start off from some place in between them, it won't be particularly clear from near the bottom which direction is more promising.

Before you get too happy with one set of results, try different guess values and try using PMETHOD=GENETIC to give a broader look at the parameter space.