I have a model which runs and simulates nicely in both Dynare and Dynare++. I’m interested in means of the stationary distribution, so I’d like to run lots of simulations to make sure I get there. The Dynare++ options ‘–rtsim’ and ‘–rtper’ (along with ‘–no-irfs’) seem tailor made for this application. Unfortunately I’m getting some unexpected errors:

I was wondering if anyone encountered a similar problem and knows what might be causing it?

The only irregularity with the model I’m using, that I can think of, other than it being relatively large - i.e. around 200 variables - is that there are some convergence issues with the steady state (so the maximum ss-tol that the model solves under is 0.00015).

The error says that the variance covariance matrix of the shocks is not positive semidefinite. This can have two causes. First is that it is really not semipositive definite (your error), second is my error. I spotted quite recently that my parser of the vcov matrix ignores signature of the numbers, so if you have some negative numbers there, then the sign is ignored. Sorry. Fix will appear in the next version.

In the meantime, if you want your shocks being negatively correlated, put it to the model, not to vcov

Thanks for the info. My variance covariance matrix consisted of just one (positive) number, so I suspect the error, this time, was not mine (I also tried a few different numbers as well, but still had no luck). I’m not sure, but to me, this doesn’t sound like a problem with the parser forgetting an odd minus (unless it creatively adds them in new places as well …) Happy to send you the model files, if that would be of any use.

Thanks again for your help (and please! don’t say sorry - the more I use dynare and realise how complex it is, the more I think you guys deserve a big monument for letting people use it free of charge).

No worries re: non-existant workaround - I’m happy to wait (particularly given that in this case I wrote a bit of code which uses Dynare++ output to compute second order accurate theoretical moments, which was all I needed really).

Anyway, great news (only n more errors remaining) & thanks a lot!
p