The RATS Software Forum

Hi,

I hope this finds you well,

I would like to implement a VAR model, where the variables experienced one sudden structural break almost in the same year.

I think that a VAR with time varying parameters is not convenient in my case, what do you think?

and if you agree with me, what model do you recommend me to apply, to model this discreate break in the context of a multivariate model?

Thank you very much for your time,

Ahmed Sahloul

If there's a sharp break in structure, it would seem that one set of estimates before and one set after would be sensible.

Would you kindly "define" sharp break in structure?

Do yon u mean instability of parameters ?

A stability test has a null that a model is stable vs the alternative that it isn't. The alternative is very broad. You'll reject stability if there is a complete break. You'll reject stability if there is a gradual but perceptible change. You're likely to reject stability if there is a permanent level shift with everything else actually being the same. You're likely to reject stability (using many tests that also assume homoscedasticity under the null) if there is a sizeable change to the variance with coefficients being the same. What you're describing sounds most closely like a complete break in the model which would require two completely independent sets of estimates.