identification problem

[ZBCPA]

Dear Johannes,

If I use data A_obs to match AR1 shock 'A' in estimation , and also add one measurement error A_me, as following

Code: Select all

A=rho*A(-1)+e_A;
A_obs=A+A_me;

Would there be identification problem, like e_A and A_me can not be identified simultaneously?

Many thanks in advance,
Huan

[jpfeifer]

No, there should not be an identification issue - unless rho=0. The reason is that e_A will introduce autocorrelation in A_obs, while A_me will not. This prevents the two shocks from being observationally equivalent.

[ZBCPA]

No, there should not be an identification issue - unless rho=0. The reason is that e_A will introduce autocorrelation in A_obs, while A_me will not. This prevents the two shocks from being observationally equivalent.

Dear Johannes,

Could I ask you another case?
If the measurement equation is

Code: Select all

A_obs=x*A;

where x is a "scale" parameter to adjust standard deviation between observed data and shock A , and x should be estimated; A here is i.i.d shock with 0 persistence, so it is just the innovation. I also need to estimate standard deviation of A. Would there be identification problem in such case?

Many thanks!

Kind regards,
Huan

[jpfeifer]

Yes, there would be a problem. You cannot estimated x and the standard deviation of A at the same time, because both capture exactly the same thing. You need to normalize one.