Measurement equation include parameters to be estimated

Dynare help (legacy posts)
1

Dear Johannes,

Could I ask you, in Bayesian estimation, if parameters to be estimated appear in measurement equation, would that be any problem?

Thanks in advance.

Kind regards,
Huan

2

Dear Huan,
why should that be a problem? A case like this occurs whenever you want to estimate the mean growth rate in an observation equation using first differences.

3

[quote=“jpfeifer”]Dear Huan,
why should that be a problem? A case like this occurs whenever you want to estimate the mean growth rate in an observation equation using first differences.[/quote]

Many thanks Johannes.

In my model , I have financial friction but do Not have a good corresponding financial data. If I use linear detrended interest rate spread data as an “instrument” , and set a new parameter measuring the volatility difference between data and model variable in measurement equation , maybe also put an measurement error there, shown as belowinterest rate spread_obs= (constant x )* model financial friction variable + measurement error ,
(model is log-linearized)
Then estimate the constant x (only appears in measurement equation) and standard deviation of measurement error, would that be a problem?

Kind regards,
Huan

4

This is hard to tell. It looks good to me. But to be on the safe side, I would test identification, if you add measurement error.

5

Dear Johannes,

Happy new year!

Could I ask further about testing identification?

  1. In Dynare, the only way to test identification is commandidentification; after estimation?
  2. If the model passes this identification check, does that mean there is No identification problem?

Kind regards,
Huan

6
  1. No, you can call identification before, after or even without estimation. All that is needed is are a varobs statement and an estimated_params block
  2. It’s somewhat more complicated. The way to check identification is local. If you pass the identification checks, that only tells you the current point is fine, but quite often that extends to (almost) the whole parameter range.