(minus) the hessian matrix at the "mode" is not positive definite

I’m getting this error with ML estimation. I know this is a common problem, but I have tried all the solutions (different starting values, different mode_compute options), and I’ve checked identification. I don’t think there’s a problem with the model: I’ve done Bayesian estimation with no obvious problem, and I’ve estimated the model with two data series and two shocks (omitting the ng series and the vn shock). When I do ML, the mode_check plots look ok, except that the standard deviation of the vu shock is implausibly large (estimating similar models it usually is around 0.2, and here it is > 1.1).

I’ve attached the .mod file and the data file, would greatly appreciate any ideas. I like to report both ML and Bayesian results, but this is a roadblock. Thank you.
Model_fttb1S3JRn.mod (8.4 KB)
cindata59gpc2024cvhp.xlsx (27.5 KB)

I used mode_compute=5 and consistently run into the parameter bounds you set. That is a typical issues with ML, the dilemma of absurd parameter values. For the start, I would try to have looser bounds and see what the implications are.

Ok, thanks. I’d tried mode_compute 3,4,8 among others and got nice-looking plots from mode_check. I guess 5 is more accurate! I’m still having trouble finding ML estimates even with widening the bounds, changing initial values, etc., but I guess that’s not unusual, even though the Bayesian results are reasonable.