EViews.com

Good morning. I have a question regarding to the especification model using Markov Switching:
1) What's the meaning of the probability regressor "c"? It is the same that the constant term traditionally named as "C"?
2) If I change the order of the regressors, I obtain different results for the coefficients and its p-values. I would like to understand the logic behind the order of the regressor in the specification of the model.
For example: my model specification is "roe pib pib(-4) gt_ap c margenfin_pat", if I change the order of "c" and the rest of the regressors, the result changes.

Thank you so much,

1. As explained in the manual it's the constant "regressor" in the transition specification. EViews allows you to add other probability regressors.

2. Markov switching models are highly nonlinear models that, by their very nature, are quite touchy. In fact, as mentioned in the documentation, they are not even really identified as I can always switch the identities of the "regimes" and results will change. More generally, changing the order of the variables will change the derivative and moment matrices in the likelihood. As with all nonlinear models, it is possible that the small differences in the order of the solutions for the moment equations will lead to path dependency. This is true for all nonlinear models but we tend not to see it. In this case, it is not a surprising finding.

How different are the results? Are they stable at the solutions? Which likelihood solution is better?

Hi there!

I am working on using the Markov Switching Model for exchange rates of Philippine Peso to US Dollar. I am studying this eviews tutorial online http://www.eviews.com/EViews8/ev8ecswit ... l#MarkovAR for reference and there are some things I do not understand. How do you determine the "regime-invariant AR(4) process" in eviews? How do you determine the probability regressors in eviews? Also, when can you tell that the markov switching model is good enough? Does the mean g have to be significant?

I hope you can help me! Thank you in advance!

Hi!
I am trying on working with a multivariate Markov Switching Model for exchange rate bubbles, especifically, with only 2 states. I am confused on how to include an independent variable to only one state (equation). Thus if I have 3 variables, one state(st=1) would have all 3 independent variables, while the other state(st=2) will only have 2 independent variables. I am guessing that maybe I can simply employ all 3 variables as switching regressors while treating one variable in state2 (st=2) as having a dummy coefficient, thus can be neglected but MS models are sensitive and I worrying that I may be doing it wrong. How can I estimate it correctly??? Please help me. Thank you.

I don't believe this can be done within EViews Markov switching feature.

The model goes like this:
Rt_1 = C + B(X1) +B(X2) + u when st=1
Rt_2= C + B(X1) +u when st=2

:( :)

You cannot impose that zero constraint in the current EViews Markov switching framework.

It's worth considering for future versions.

Being able to put in constraints in lots of places would be worth considering for future versions. :D

Being able to put in constraints in lots of places would be worth considering for future versions. :D

Think about the unintended consequences if I directly interrupted your work day with questions even more often!