Dynare vs. dynare++ results

Dear all,

I encountered a problem when comparing the results of Dynare and Dynare ++ and I would like to ask whether these problems are known or not and what would be your suggestion….

The problem arises to some extent already when running a simple RBC model. While the IRFs produced by Dynare and Dynare++ look the same, the steady state (fixed point) and policy functions are different after the 4th digit (comparing the results of the 2nd order approximation of the model).

The problem becomes even bigger if I allow for the variance of the technology shocks in my model to be time varying and follow an AR(1) process. I make this modification to the simple RBC .mod file and when comparing the results of Dynare and Dynare ++ (2nd order approx.) I again get visually similar IRFs (but not of exactly the same size) and again the stochastic steady state and the policy functions are not the same after the 4th digit.

When I go for higher orders of approximation in Dynare++ the IRFs for the shocks to variance change considerably, especially if I allow for a big uncertainty shock (in that case they sometimes even change sign).

Going for a more complicated model with time-varying variance makes things even worse. Now, the IRFs are of the wrong sign (comparing to the Dynare output) already at the 2nd order approximation.

Is there any known reason for these differences? Which routine should I rely on in case of having time varying variance of the shocks in the model?

Maybe it is of some importance that I wrote all the models with variables in logs and not in levels.

I would really appreciate your help.



Does your model contain leads of 2 or more?

We discovered a bug in Dynare 4.0.4: it gives wrong results when there are leads of 2 or more on endogenous variables. So if this is your case, you probably should trust Dynare++.

This bug is now fixed in the development version of Dynare, and will be fixed in the upcoming 4.1 release.