System hangs during Dynare++ call

[amfasolo]

Hi everyone,

I am running simulations using Dynare++ to solve the model, but I have an issue I can't find a solution. Sometimes, when Blanchard-Kahn conditions are not satisfied with the parameters of the simulation, the system hangs, instead of exiting Dynare++ execution with an error call. I'm running the simulations in a 64-bit computer with Windows Professional edition, Matlab version R2015a.

In order to provide a summary of the code, the simulations runs as follows:
1) choose a set of parameters to simulate the model;
2) run Dynare++ with a call given by: "checksol = system('C:\dynare\4.4.3\dynare++\dynare++ --order 3 --no-irfs --no-centralize model.mod');", where checksol is the outcome of the call to "system" (result is zero if everything is fine).
3) given the solution of the model, run simulations; otherwise, back to step 1;
4) store simulation results, back to 1.

The whole issue is during step 3, when the call to the command in Matlab simply does not return. If I interrupt the execution, I can see that the journal file (.jnl file produced by Dynare++) finished the execution and produced an outcome saying that the system did not satisfied the Blanchard-Kahn conditions. Another interesting point: if I call the "system" command again, right after interruption, I get a normal error output, as expected.

Any help would be greatly appreciated.

Best,

Angelo

[jpfeifer]

Maybe this is a stupid question, but considering that you only use order=3 and the no-centralize option, why are you using Dynare++ in the first place? The simulation setup you describe can be easily handled by Dynare.

[amfasolo]

I use the policy functions to compute a likelihood function as well. Thus, not only I run a third-order approximation, but also a linear approximation (for pruning purposes) and run all the simulations and compute the likelihood as a final outcome of each step.

[jpfeifer]

Those are also all things Dynare can do (including the pruning). What exactly is it that requires Dynare++?

[amfasolo]

Dynare now estimates models using third-order approximation of the equilibrium conditions as well? Cool. I knew only about the second-order solution. Does it run also altetnative procedures to the simple particle filter (the bootstrap filter)? Unfortunatelly, the bootstrap filter does not perform really well with data with large volatility, requiring an unfeasible number of particles for stability.

[jpfeifer]

Not completely. The particle filter is still experimental and only works up to second order currently. But at order=3, Dynare can do everything that Dynare++ also does. For simulations, it is even more convenient, because the simult_ function allows you to easily simulate data using a (pruned) third-order approximation.