Page 1 of 1

Problems with forecasting

PostPosted: Fri Sep 06, 2013 1:21 pm
by ger163609
Hello!

I have some troubles with two of the options in the estimation command, namely the forecast=PERIODS option and the conditional_variance_decomposition=[PERIODS] option.
Everything works fine. After running the MH algorithm the MCMC diagnostics and the posterior simulations are spitted out and they look ok. However, afterwards the estimation cannot be completed, i.e. the forecasts and the variance decomposition do not pop up. Instead, the following error message appears:

Error using chol
Matrix must be positive definite.
Error in conditional_variance_decomposition (line 56)
B = StateSpaceModel.impulse_matrix*chol(StateSpaceModel.state_innovations_covariance_matrix)';
Error in dsge_simulated_theoretical_conditional_variance_decomposition (line 124)
Conditional_decomposition_array(:,:,:,linea) = conditional_variance_decomposition(StateSpaceModel, Steps, ivar);
Error in posterior_analysis>job (line 75)
[nvar,vartan,NumberOfFiles] = ...
Error in posterior_analysis (line 34)
oo_ = job(type,SampleSize,arg1,arg2,arg3,options_,M_,oo_);
Error in compute_moments_varendo (line 116)
oo_ = posterior_analysis('conditional decomposition',i,M_.exo_names(j,:),Steps,options_,M_,oo_);
Error in dynare_estimation_1 (line 941)
oo_ = compute_moments_varendo('posterior',options_,M_,oo_,var_list_);
Error in dynare_estimation (line 70)
dynare_estimation_1(var_list,dname);
Error in NKM_cor_bm_est_01_08_modeified_priors (line 524)
dynare_estimation(var_list_);
Error in dynare (line 120)
evalin('base',fname) ;

After thoroughly revising the forum this has to do with the mode computing technique. I am using mode_compute=9, but I also tried mode_compute=8 and the problem is still there.

Can someone help me on this?

Thanks a lot in advance!

Re: Problems with forecasting

PostPosted: Fri Sep 06, 2013 2:19 pm
by jpfeifer
This might be related to the bugfix at http://www.dynare.org/phpBB3/viewtopic.php?f=1&t=4961

Re: Problems with forecasting

PostPosted: Fri Sep 06, 2013 3:39 pm
by ger163609
Thanks jpfeifer!

You wrote in the other thread that the problem was because of the variance of one shock set to zero. This is not the case in my model, as I estimate all of them.

Anyway, I will try it and report the results!