Bayesian estimation/forecast issue after large MH draws

I’m having trouble in estimation of a small open economy model. I think the problem is with the forecast command, because the error occurs right after the print of the point/mean forecast graphs.

model_diagnostics command indicates that there is no problem with the model, and i had run the code twice with fewer iterations with no problems (except the fact that there was convergence issues). More specifically, i’ve used the load_mh_file for the third time in order to achieve convergence in a total of 4600000 MH draws per chain (4 chains), and then this error appeared, just when the MH seems to have converged.

I’ll attach the mod file (w/ mh_mode + data file ) but i dont think it is feasible to reproduce what i’ve done to get the error, so i hope that anyone can help without executing the mod file.

Error message:

Error using print_info (line 32)
Blanchard & Kahn conditions are not satisfied: no stable equilibrium.

Error in DsgeSmoother (line 108)
print_info(info,options_.noprint, options_);

Error in dynare_estimation_1 (line 558)

Error in dynare_estimation (line 105)

Error in WT.driver (line 665)

Error in dynare (line 294)
evalin(‘base’,[fname ‘.driver’]) ;

data2015tvit.mat (3.0 KB) WT.mod (13.8 KB) WT_mh_mode.mat (16.3 KB)

This is tricky. Your code crashed during the run of the classical smoother, although you ran an MCMC. I guess you did not run additional replication the last time. That may explain the issue. You will end up with the parameter vector being set to the posterior mean. But there is nothing that guarantees the posterior mean to be a feasible draw as it is an average.

I did run additional replication (2 million) last time.

Which Dynare version were you using? Also, did you have a look at the trace plots for your chain?