Collinearity and Model improvement

Greetings,
I will be grateful getting your advice on ways to improve my model. I spent several hours going through similar questions on the forum but still not clear.

I am following this paper with a bit of twist from C. Junior. I am able to reach the SS but facing two issues :

  • Blanchard & Kahn conditions not satisfied, in order 2, even after setting qz_zero_threshold to 1e-30. The estimation works using order 1. But I am planning to introduce welfare analysis
  • When I go for order 1 I am able to run but I see NaN in most theoretical moments, and unit roots and collinear equations issues after running model_diagnostics(M_,options_,oo_).

I read that a way to go about it in a NK model is to reestimate the model in terms of inflation instead of price level which is non-stationary. The problem is that the model is estimated in real terms (as in the paper) and none of the equations I am replicating is expressed in terms of inflation. Which makes me believe that the authors were able to work this out. Also, I have not seen a similar approach in C. Junior.

Please correct me if I am wrong and advise on ways to improve my model . Thanks a lot

Food subsidies

Model_Dyn.mod (10.7 KB)

You did not provide all files to run the model.

Sorry @jpfeifer
script_v13CNR_steadystate_helper.m (252 Bytes) script_v13LMDR_steadystate_helper.m (270 Bytes) script_v13LNR_steadystate_helper.m (303 Bytes)

Are you sure your steady state makes sense. E.g.:

LAMBDAR || 9.31329e-12
lambda  || 9.31329e-12

being pretty much 0 is strange.

Thanks. They are both derived from the _steadystate_helper. I am trying to work on reducing the number of equations and getting rid of some variables including lambdas (to the extent possible). I do not know whether that could help addressing with the super high eigen values. But I will see.

  1. The problem is not the super high eigenvalues.
  2. It’s not yet clear whether there is a fundamental problem in your model like a missing equation or whether the different size of variables in steady state causes a numerical problem. That’s why I was asking. An almost 0 value for a variable that cannot be 0 is always suspicious.

Thanks a lot. I am restarting from scratch excluding government and subsidies to spot possible errors.