Two potential problems about Dynare and the Reference Manual
Posted: Tue May 31, 2016 3:45 amMy version is 4.4.3
1. It is said in page 63 of Reference Manual:
"After estimation with Metropolis iterations (option mh_replic > 0 or option load_mh_file set) the parameters M_.params and the variance matrix M_.Sigma_e of the shocks are set to the posterior mean."
No. I use the following command:
estimation(datafile=mydata_datafab_lev,mode_compute=4,mh_replic=2000,mh_nblocks=2,mh_drop=0.45,mh_jscale=0.6);
But "M_.params" just stores the mode.
This may lead to another problem. That is, if I try to plot IRFs by adding the following command:
stoch_simul(conditional_variance_decomposition=[1 2 4 8 16]) c i y lab u ele w GR pi;
It is in fact the mode of the parameters used. I compare this plot and that generated by a model given parameter values as the posterior mean, there are different. However, if I let the parameter values as the posterior mode, they are the same.
2. About historical and smothed variables
I estimate a model without measurement errors and stochastic singularity with the following command:
varobs y_obs i_obs pi_obs;
estimation(datafile=mydata_datafab_lev,mode_compute=6,mh_replic=10000,mh_nblocks=2,mh_drop=0.45,mh_jscale=0.6,moments_varendo,selected_variables_only,conditional_variance_decomposition=[1 2 4 8 16]);
A problem arises: Historical series of pi_obs is far different from its smoothed series.
The two become the same if I delete the last three options in the estimation block and use the following command:
estimation(datafile=mydata_datafab_lev,mode_compute=6,mh_replic=10000,mh_nblocks=2,mh_drop=0.45,mh_jscale=0.6)
1. It is said in page 63 of Reference Manual:
"After estimation with Metropolis iterations (option mh_replic > 0 or option load_mh_file set) the parameters M_.params and the variance matrix M_.Sigma_e of the shocks are set to the posterior mean."
No. I use the following command:
estimation(datafile=mydata_datafab_lev,mode_compute=4,mh_replic=2000,mh_nblocks=2,mh_drop=0.45,mh_jscale=0.6);
But "M_.params" just stores the mode.
This may lead to another problem. That is, if I try to plot IRFs by adding the following command:
stoch_simul(conditional_variance_decomposition=[1 2 4 8 16]) c i y lab u ele w GR pi;
It is in fact the mode of the parameters used. I compare this plot and that generated by a model given parameter values as the posterior mean, there are different. However, if I let the parameter values as the posterior mode, they are the same.
2. About historical and smothed variables
I estimate a model without measurement errors and stochastic singularity with the following command:
varobs y_obs i_obs pi_obs;
estimation(datafile=mydata_datafab_lev,mode_compute=6,mh_replic=10000,mh_nblocks=2,mh_drop=0.45,mh_jscale=0.6,moments_varendo,selected_variables_only,conditional_variance_decomposition=[1 2 4 8 16]);
A problem arises: Historical series of pi_obs is far different from its smoothed series.
The two become the same if I delete the last three options in the estimation block and use the following command:
estimation(datafile=mydata_datafab_lev,mode_compute=6,mh_replic=10000,mh_nblocks=2,mh_drop=0.45,mh_jscale=0.6)