Computing the mode

Estimation ML/Bayesian estimation
1

Hi!

I’m trying to estimate a model, but I’m stucked in finding the mode. I’ve gone through all options of mode_compute, but in all I get an error saying that the hessian is not positive definite. The only ones that seems to work is mode_compute=2, 6, but the final results shows poor convergence (this version only has 100k, but I’ve tried with 3M and it doesn’t seem to converge).

I have 3 questions: 1) what could I do to compute the mode? Is this, most likely, caused by model misspecification? problems with the data?
2) I read in the book by herbst and schorfheide (2016) that when there are problems computing the mode, it might be a good idea to use different methods. Dynare has ‘HSSMC’ option, but would be a viable alternative? Is there a rule-of-thumb for the number of steps and particles that I would need to use?
3) In case of using mode_compute=6, should I try to estimate really long chains? Say, for example, 10M?
4) Should I be worried about the other methods not computing the mode?

Thanks!
data_use.mat (13.6 KB)
steady_modelV6.mod (14.1 KB)

2

Investigating the mode_check plots shows that rhoA moves towards a corner solution of 1. The need for a unit root is usually caused by the mean of model variables not matching the data mean. Usually that’s the type of issue you need to fix for any approach to be successful.

3

Thanks a lot!!