Problem in deterministic simulation

[ripatti]

Dear All,

I experiment a deterministic simulation in with my model. I change the endval one of my variable (by 1 %), and obtain the following error:

Code: Select all

-----------------------------------------------------
MODEL SIMULATION

variable SRD       period 2
1-   err = 13.9451
err_f = 22.9525
   Time of this iteration   : 8.695
variable RSR       period 6
2-   err = 4395.4199
err_f = 30.4353
   Time of this iteration   : 8.835
variable ERSW      period 2
3-   err = 341628173856777470
err_f = 1459996.1773
   Time of this iteration   : 23.06
Warning: Matrix is close to singular or badly scaled.
         Results may be inaccurate. RCOND = 3.050461e-036.
> In simk at 206
  In simul at 60
  In Aino at 1620
  In dynare at 79
Warning: Matrix is close to singular or badly scaled.
         Results may be inaccurate. RCOND = 1.792819e-036.


These error messages keep on coming until MaxIters is reached. Is there any way to get more information what might be the problematic part of the model code?

I can make such a permanent shock with another model variable without problems. The same error message is present also in the dynare version 3.

Best regards,

Antti

[ripatti]

I found some information
http://www.dynare.org/manual/re25.html *

The symptoms in my case a similar of those.

Sorry for not looking carefully the forum before posting.

Antti

NB: * new ref

[MichelJuillard]

In this case, increasing the number of iterations won't help. As you can see for the size of the errors the Newton procedure explodes instead of converging. It's because the variables explode that a matrix ends up singular.

Try to diminish the size of the shock and see if you can solve it. Then you can try to increase the size the shock in a loop using the previous solution as initial value for the next experiment.

Best

Michel

[ripatti]

Many thanks Michel!

Does Dynare 4 automatically use the previous solution as the initial value? Ie, should the following code do the job?

Code: Select all

for step_ = 0.996:-0.001:0.99;
   endval;
      UPSILONSS = 1.085*step_;
   end;                         
   steady(solve_algo=3);
   disp(['Shock size: ' num2str(step_)]);
   simul(periods=200);
end;   


Best
Antti

[MichelJuillard]

Yes, it should. Does it?

Best

Michel