Overidentified SVAR with Short- and Long-Run-Restrictions

[jonasdovern]

Hello,

I am wondering how to estimate an SVAR using short-run and long-run restrictions that has more restrictions than necessary to just-identify it.

In particular, I am thinking about a 4-variables model with two shocks that have no long-run effects on two of the variables, and one of these two shocks plus one of the other two shocks without any restricted long-run effects shouldn't impact two of the four variables contemporaneously.

Using the usual notation used e.g. for @shortandlong I thus have:

Code: Select all

dec rect lr(4,4) sr(4,4)
input lr
 0 0 . .
 . . . .
 0 0 . .
 . . . .
input sr
 . 0 . 0
 . 0 . 0
 . . . .
 . . . .

Since @shortandlong cannot deal with overidentified SVARs, my question is, how I can estimate such a model using cvmodel.

[TomDoan]

This is addressed in the RATS v8 User's Guide on page UG-221. In effect, you can't because the overidentification places restrictions on the lag coefficients.

[jonasdovern]

Thanks for pointing to the paragraph that I missed in the User Guide.

I guess the "rather nasty non-linear constraints" are either very hard to derive and/or prohibit a proper convergence of the estimation procedure?

[TomDoan]

They would be implicit functions of the dynamic behavior of the whole model. I've never seen an example of anyone doing it, so I have no idea whether it's computationally feasible.