Dynare forums

We use the latest stable version of dynare (4.4.3) and try to produce a deterministic simulation with the option "block". When we use the stack_solve_algo from 0 to 4, we get the error message: Undefined variable "options" or class "options.simul.maxit". This happens even when we do not use the maxit option; and any attemp to introduce a maxit definition fails. What can we do? Thanks in advance for any help.

PS: We could obviously run the simulation without "block" and "stack_solve_algo=0/6" but the model size is expected to increase and the block option could also become helpful.

You should post your code and source material (citations of similar projects, etc.) as well, it'll be difficult for someone who knows the relevant information to diagnose your problem without being able to look at what's going on and the source of the project.

You'll find below a zip folder including the mod file (loading a mat file containing initial and final steady state values). As said in my previous post, I have a problem when I run it with stack_solve_algo from 0 to 4 and the block option. This model relates to the paper I also attach.

This is a bug in Dynare 4.4.3. Please use the unstable version. There, the problem is already fixed. The issue is that the preprocessor misses an underscore when creating the call to solve_one_boundary

Thank you Johannes. We can now simulate the model with the unstable 4.4.4 version with the block option (though it does not converge with the block and bytecode options together).

However, we have another problem. When we take exactly the same OLG model but with a 1-year generation instead of a 5-year generation (basically we multiply the size of the model by 5), we are unable to obtain convergence anymore when simulating the model, whatever algorithm and options (no option/block/bytecode/ block&bytecode).

Would you have some hints? Is the model too large for the dynare solver? Is the model too large for our computer (all CPU is utilized during the simulations)? Are there some tricks we could use as reducting the number of periods? Is there algorithm/options we should prefer? Are the shocks too large (but the shock sizes are the same than in the 5-year version)?

As before, you'll find the mod file and the mat file (containing initial and final steady state values, uploaded by the mod file).

Sorry for these -- very -- general questions but we do not know in which direction we have to go.

This is not really my hometurf. But your model looks really huge. Have you checked whether the exogenous processes stored in

Code: Select all

oo_.exo_simul

make sense? It seems that the steady states used for endval imply a strange jump at the end.

We have been working on it these days and your guess is fully right. We had a problem with the final steady state value related to the foreign population shock. After correction, the deterministic simulation (although the model is huge) easily converges. Thanks again for your help.

Dear all,

I also have the following report:

==== Identification analysis ====

Testing posterior mode

All parameters are identified in the model (rank of H).

WARNING !!!
The rank of J (moments) is deficient!

==== Identification analysis completed ====

0.6% of the prior support gives unique saddle-path solution.
59.5% of the prior support gives explosive dynamics.
0.7% of the prior support gives indeterminacy.
For 39.2% of the prior support dynare could not find a solution.


Smirnov statistics in driving unique solution
c2rhoomegacm2 d-stat = 0.642 p-value = 0.000
c2gamdpc d-stat = 0.582 p-value = 0.000
c2rhogy d-stat = 0.551 p-value = 0.000
c2rhozi d-stat = 0.539 p-value = 0.001
c2rhomuoc2 d-stat = 0.522 p-value = 0.001


Correlation analysis for unique Stable Saddle-Path
No correlation term with pvalue <1e-05 and |corr. coef.| >0 found for unique Stable Saddle-Path

Correlation analysis for NO unique Stable Saddle-Path
No correlation term with pvalue <1e-05 and |corr. coef.| >0 found for NO unique Stable Saddle-Path

Smirnov statistics in driving indeterminacy
c2rhomuoc2 d-stat = 0.542 p-value = 0.000
c2rhoomegacm2 d-stat = 0.520 p-value = 0.000
c2delta d-stat = 0.520 p-value = 0.000
c2rhogy d-stat = 0.512 p-value = 0.000
c2gamdpc d-stat = 0.491 p-value = 0.001


Correlation analysis for NO indeterminacy
No correlation term with pvalue <1e-05 and |corr. coef.| >0 found for NO indeterminacy

Correlation analysis for indeterminacy
No correlation term with pvalue <1e-05 and |corr. coef.| >0 found for indeterminacy

Smirnov statistics in driving instability
c2delta d-stat = 0.966 p-value = 0.000


Correlation analysis for NO explosive solution
No correlation term with pvalue <1e-05 and |corr. coef.| >0 found for NO explosive solution

Correlation analysis for explosive solution
No correlation term with pvalue <1e-05 and |corr. coef.| >0 found for explosive solution

Smirnov statistics in driving inability to find a solution
c2delta d-stat = 1.000 p-value = 0.000


Correlation analysis for NO inability to find a solution
No correlation term with pvalue <1e-05 and |corr. coef.| >0 found for NO inability to find a solution

Correlation analysis for inability to find a solution
No correlation term with pvalue <1e-05 and |corr. coef.| >0 found for inability to find a solution

Computing theoretical moments ...

... done !

==== Identification analysis ====

Testing posterior mode

All parameters are identified in the model (rank of H).

WARNING !!!
The rank of J (moments) is deficient!

Monte Carlo Testing

Testing MC sample

All parameters are identified in the model (rank of H).

WARNING !!!
The rank of J (moments) is deficient for 9 out of 13 MC runs!

==== Identification analysis completed ====[/quote]

Sample check OK 0
Total computing time : 4h02m34s

Is the problem related with the steady state or from some kind of misspecification on the model code?