Unit roots, unity eigenvalues

[martom]

Hello,

After a thorough use of Dynare - and our delight with the program - we are running into some difficulties, most likely tied to our model.

Everything worked fine until we made a minor change; the results we are now getting don’t make any sense anymore.

We suspect one thing: we’re getting an eigenvalue of one.

1. how does Dynare deal with this? Is this a problem?

2. how do we know to which variable this eigenvalue corresponds?

We are also getting a value of one in a coefficient of the policy and transition function.

3. is that a problem?

We don’t think so, but it’s probably worth checking.

We’re under a pretty tight deadline to submit this paper, so any help would be immensely appreciated.

Thanks!

[MichelJuillard]

@martom

We suspect one thing: we’re getting an eigenvalue of one.

1. how does Dynare deal with this? Is this a problem?

one is still treated as a non-explosive root. The model is stable around a stochastic trend.

2. how do we know to which variable this eigenvalue corresponds?

when you run stoch_simul with version 4. The unit root variables should disapear from the list of moments because they one can’t compute their variance. (OK, it would be better to report them with infinite variance).

Be carefull this is broken in version 3 that can report finite variance for unit root variables!

We are also getting a value of one in a coefficient of the policy and transition function.

a coefficient of one isn’t necessarily meaningful, what matters are the roots of the transition matrix as a whole

3. is that a problem?

It isn’t a technical problem. But I think it is a potential model problem to have a unit root in the model and not to know its origin.

Best

Michel

[martom]

Thanks so much for these precious and particularly prompt remarks. We will immediately work with them to try to improve our results.

A very helpful newsgroup indeed