Number of obs variables vrs number of shocks

Estimation ML/Bayesian estimation
1

Hi,

I am estimating a model with 40 equations, 30 parameters and 7 variables. I also have 10 shocks. I read from this platform that the number of shocks must be equal or more than the number of observable variables. However, I also read that too many shocks that are not linked to the data may cause identification problems. I am not sure to what extent can the shocks be more than the variables.

In my case, there are 3 more shocks than the variable. Will those pose shock identification problems as there are many shocks that are not matched with the data? If you think the shocks are too many, what will be the most appropriate way to handle it? Drop some shocks or calibrate some shocks.

If I am to calibrate the shocks in a model where I am estimating some parameters, how do I go about it? Assuming I have two shocks and I would like to calibrate one (eps_u) and estimate the other (eps_z) Should I use:

shocks;
var eps_u = 0.5;
end;

estimated params;
stdeer eps_z, inv_gamma_pdf, 0.5 , inf;
end;

Lastly, and most important, when conducting optimal monetary policy analysis with NK model, is it ok to have more shocks than variables in the model? I am using the estimate command.

2
  1. There is no general rule, but the identification-command allows testing for issues.
  2. If there is a problem with identification of shock persistences and variances, you can indeed calibrate them. The syntax above is correct.
  3. It does not matter if you are doing an optimal policy exercise. That does not alter anything.
3

Your response clears up a lot of confusion in my mind. I have problems with identifying a few shock persistence. I will calibrate them as suggested.

Enjoy your weekend.