Hi everyone,
I download the new stable version 4.5 and I tried to run an old mod file that was created under version 4.4.3
With the older version the estimation was fine while when I run the mod file with the new version 4.5
after the mode search (I used mode_compute=5, but the results does not depend by the algorithm) I obtain the usual warning that the variance and covariance matrix is not semidefinitve positive and the estimation collapse.
Minimum and Maximum eigenvalues are quite different between the two versions
I check the updates list but it seems that you did not intervene on this part of dynare.
Do you have any idea what can be the problem here?
Thank you
Federico
Have you compared the mode found by the two versions? Is it the same?
Yes, the mode found with version 4.4.3 and 4.5 are the basicly thesame but I cannot obtain the standard deviation
However, If I used othe options to compute the hessian at the mode and not the default option (‘Hessian’, 2) I was able to obtain results comparable with the older version.
In this case the convergence is harder to achieve for a couple of parameter.
Can it be a solution ?
I think there has been a change in the way the Hessian is computed with mode_compute=5. That would explain the difference and why using a different way to compute the Hessian is successful.
The only strange thing is that with the default option
I obtained an hessian matrix where an entire row is a NaN.
That’s why of course the hessian is not invertible.
Could you please provide me with the files to replicate the issue. There may be a problem with Dynare.