Dynare forums

Hi all,

To bad that my first message here is asking for help. I have this model which is somewhat simple (only 10 equations, 10 unknowns), but I am not being able to get Dynare to converge. In fact, right now I am giving dynare the stationary steady state values obtained by using fsolve and still no luck. I have played with parameters thousands of times, and nothing. Perhaps you guys here can have some advice for me. I am attaching so you can see the model and the error.

Thanks for your time

Your starting values are really far away from the true steady state. You probably need to compute the steady state by hand with pencil and paper (at least partly).

@jpfeifer, thank you very much for your response. I was working on those initial values. Now, the residuals seems to be closer to zero, but the issue is that now the program works in some computers and in others it just does't. Why do you think could be the cause for that?

Besides, even in the computers in which the program is able to run, that outcome is extremely sensitive to the parametrization.

Right now in the computer I am, it seems that is getting the steady-state values but I get

Code: Select all

??? Error using ==> print_info at 42
Blanchard Kahn conditions are not satisfied: indeterminacy

Any advice on how to proceed from here? I am attaching my code so you can see for yourself.

Again, thank you so much.

You entered your SS with fixed numerical values, but the SS is a function of the deep parameters. Hence, if you change the parameters, the SS changes but not your initial values, which the become poor starting values. Regarding the BK conditions, search the forum for a million similar questions. Most commonly it is an issue of the timing of capital or bonds, which are predetermined states (see the Dynare manual).

I understand that, problem is, even though the steady state almost doesn't change at all (according to the result from fsolve), dynare is not able to solve for the new parameters.

I checked the timing convention and it seems to be right.

I would recommend writing a steady state file that supplies the starting values for a given set of parameters. There are quite a few posts on how to do that.

What about for example the first equation? If

Code: Select all

(exp(k)/exp(lab) )^alpha

is the capital to labor ration, you have lab in period t and k in t+1, if k is a predetermined state.

Thank you very much jpfeifer. That was a silly (to say the least!) mistake.

One more question: I have 14 variables, however, the matrix oo_.dr.ghx is 15x6 so it is basically adding one more variable. Besides, when I look at its values, one entire raw is full of zeros. Do you know why that could be the case?