Dear Michel,

One more question about osr in Dynare and under a non-linearized model. Searching for the optimal parameters in the rule

Rule 1: i = -(1-para_i)log(beta) + para_ii(-1) + (1-para_i)para_yygap + (1- para_i)para_pieinflation

returns a value para_i smaller then, but very close to 1. Consequently, the overall weight on output and inflation goes to zero (the values of para_pie and para_y are around 2 and 1 respectively).

Now using the rule

Rule 2: i = -(1-para_i)log(beta) + para_ii(-1) + para_yygap + para_pieinflation

i.e. para_pie instead of (1-para_i)*para_pie and para_y instead of (1-para_i)*para_y returns the error message for non-convergence:

??? Error using ==> steady_
STEADY: convergence problems

I cannot see where the difference comes from, i.e. why osr has problems with rule 2 but not rule 1. I did even scale the initial parameter values in rule 2 so that the overall coefficients on inflation and the output gap are identical in both rules.

The non-convergence message for rule 2 even appears if I fix the parameter for interest rate persistence, para_i, to say 0.9 and than optimize over para_pie and para_y only.

Do you have any idea, where this difference between in principle identical rules may come from?

Many thanks!


Dear Lukas,

the problem is about the steady state of the entire model, not the dynamic convergence.

When you change the specification of the money rule, you change the Jacobian of the model and the Newton method inside STEADY maybe have on principle more difficulties. But, it shouldn’t happen in practice.

Which version of Dynare are you using?

Can you send the *.mod files?