Is it possible to reuse preprocessing output?

#1

Hello,

I am new to dynare. I have a model with 200 variables and many more model-local variables (about 2000). Preprocessing this on my windows pc with 8gb ram takes close to 2 hours. Is it possible to store the output of the preprocessing step and use it in the next run of the model?

Thank you

#2

That depends on what you mean with

What exactly is the experiment you have in mind? Preprocessing is required whenever you change the structure of the model.

#3

I am running stoch_simul and am using a Matlab file to get the steady state. To start execution, I am running eval([‘dynare dynamicModel.mod’]) in Matlab. When I say “next run”, I mean running this eval statement again.

If I do not change the structure of the model or the values of the parameters, does the preprocessor automatically understand that it can use the output from the previous run or do I have to ask it to use the last output with some command?

Thank you

#4

That does not really answer my question. Most of the times, there is no purpose to rerun Dynare on the mod-file. It is often sufficient to call the particular Matlab functions of Dynare again. That even allows changing parameters. Examples of this are estimation via moment or IRF matching.

#5

I see. I was not aware of this. I am trying to generate IRFs to the steady state of my model. Which Matlab file should I run the next time once preprocessing is done? Does it have a specific name?

#6

You still did not answer the question what you are changing between different runs of the model. Without knowing that, it is impossible to tell.

#7

Sorry, I am only changing a few parameter values between different runs of the model.

Can I change that in some file generated by the preprocessor and run that file instead of rerunning the preprocessor? Which is that file?

Thanks a lot!

#8

In that case, you run the model once and after that only call the main m-file created by the preprocessor. After that, you can use set_param_value to change parameters and can call functions like stoch_simul directly. See e.g.

1 Like