I am doing a bayesian estimation and I end up with more shocks than observable variables, I would like to know if this is an issue.
Normally the problem is the opposite, when you have less shocks than observable you have singularity and the estimation is not feasible.
In my situation I think to have two options. The first is to estimate a number of shocks equal to the number of observables while the exceeding shocks are not estimated and simply declared in the estimation mod file as:
- Code: Select all
model (linear)
x=rhox*x(-1)+ex; //stochastic process
...
shocks;
var ex; stderr 0.01;
....
The other option is to remove the whole stochastic process relative to the exceeding shock (in the above example x). In this case the stochastic variable becomes a simple parameter of the model but his would mean to re-linearize the model treating the former stochastic variables as parameter (much more work).
Which of the two is the correct way of dealing with this issue?
Thanks a lot!