If the number of shocks is less than the observables(US data) in Bayesian estimation, for example, 5 shocks but 7 observables,
1.If I add two measurement errors, is there any drawbacks to do so?
2. Is it OK to add 7 measurement errors (like real business cycles in emerging countries)?
Many thanks in advance,
- As long as all parameters are still identified, there is no problem. Many people would argue that this is preferable to droppping observables.
- That would be OK, but increases the parameter space considerably and is therefore not done that often.
I personally would go for option 1, putting measurement error on e.g. wages or investment.
Many thanks for your very helpful reply!
Could I ask again how to make sure that all parameters are all identified? See posterior plots?
You can either run the identification command if the manually added the measurement error as “structural” shocks using varexo. Or you can look whether the mode_check-plots shock horizontal lines. However, the latter way would not detect collinearity between parameters.