Mode check

[jmaih]

Hi,
Is it normal that some of the endogenous variables appear among the list of the parameters for which the mode is being checked? In the example below 'AUX_YH', 'BSTAR' and 'C' are endogenous variables and the other elements are parameters.


\nMODE CHECK\n
Fval obtained by fmincon: 0.000000
'sig_Zc'
'sig_Zls'
'sig_ZpiT'
'sig_Zpizy'
'AUX_YH'
'BSTAR'
'C'
'alph'
'delt_f'
'ups'
'vsig'
'vrhoxp'
'vrhoxy'
'rho_pit'
'rho_thet_h'
'rho_thet_w'

Most negative variance -319132.408265 for parameter 34 (rho_thet_w = 0.997637)

Also, what does "Most negative variance" mean?

Junior

[MichelJuillard]

Dear Junior,

the endogenous variable names most likely refer to the standard error of measurement errors associated with this variable. Please tell me if it isn't the case.

Theoreticaly a variance can't be negative and the Hessian matrix is definite positive, but, if the optimization isn't successful and stops away from the mode, the Hessian matrix won't be definite positive. It is possible to apply a correction to make it definite positive, but, in that case, it is crucial to know the importance of the correction.

Dynare reports the most negative variance before the correction. In your example, this is a very big negative number and something very wrong seems to have happens during optimization.

Kind regards

Michel

[jmaih]

Dear Michel,
Thank you for the prompt reply and for the clarifications. Unfortunately the variables appearing in the list are unobserved and therefore do not refer to measurement errors. However, I do have 3 measurement errors in the model. I can send you the files on request.

Thanks,
Junior

[MichelJuillard]

Yes please, send me the files. It looks like there is a bug in the labeling of the variables

Best

Michel