Hessian matrix at the "mode" is not positive definite...but the marginal density is displayed anyway?

Good afternoon,
During an estimation, I am getting the error message below. My understanding from reading the posts on this topic is that this means (or likely means) that the point say fmincon converged to is not in fact the mode. I have one clarification question: What if dynare displays this error message, but still reports values for all the standard deviations of the estimated parameters (instead of ‘NaN’), as well as the la place approximation of the marginal density. I thought the la place approximation and the standard deviations of the estimates and are based on the Hessian…? Could this mean that the error message is a bug in this case?
Many thanks,
Ansgar

POSTERIOR KERNEL OPTIMIZATION PROBLEM!
(minus) the hessian matrix at the “mode” is not positive definite!
=> posterior variance of the estimated parameters are not positive.
You should try to change the initial values of the parameters using
the estimated_params_init block, or use another optimization routine.

That is strange. Could you please provide me with the mode-file?

Hi Johannes,
thank you for your quick response!
I am attaching the _mode file and the log file.
Edit: I just realized I was using dynare 4.5.1, so I redid the estimation and replaced the _mod and log file. The issue is there in 4.5.7 as well.
Best,
Ansgar

Exe_OISEONIAnewNDM2_theta096_mode.mat (21.7 KB)Exe_OISEONIAnewNDM2_theta096.log (30.1 KB)

This is indeed a bug. The determinant is negative, so the log determinant becomes complex. But the display command in Matlab only displays the real-valued parts. Can you confirm that oo_.MarginalDensity.LaplaceApproximation is indeed complex?

Hi Johannes,
please excuse the long delay in replying! Yes it is indeed complex: -472.580021437163 + 4.71238898038469i

Best,
Ansgar