Question about numerical solution and Bayesian estimation?


#1

Dear Professors,

I have ecountered a problem when I wanted to do the Bayestion estimation. Would you please give some advice? Thank you very much !

Problem description:
Some variable’s steady state in my model does not have a analytic solution, so I have to use the numerical solution. However, in the process of finding the numerical solution, I should use the parameter (e.g. phiUU) that will be estimated. So if this phiUU change during the estimation, the variable’s steady state will change as well. But as it is in numerical form, it could not change during the estimation.
In this case, how could I run the Bayesian estimation?(associated documents are attached)

Any reply will be appreciated. Thank you!


#2

This is a straightforward example of parameter dependence. You need to write a steady_state_model-block or a steady state file where you call the numerical solver. That makes sure that the parameter updating is correct.


#3

Dear Professor Pfeifer,

I wrote a steady state file which runs successfully in the model without Bayesian estimation block. But when I try it in the model with Bayesian estimation, it says ‘‘The steadystate file did not compute the steady’’. (attachments are mod file, steadystate file and data file)

What may be the problem?

Any reply will be appreciated. Thank you !


#4

In your steady state file, you have

RV   =log(RVV);
...
RV_ob=0;

while in the model

RV_ob=RV;

So unless RVV is always 1, this cannot be true.