Stoch_simul from recursive estimations

Stochastic simulations
1

Dear all,

I have recursively estimated my model and now I am trying to load all the recursively estimated parameters and run a stoch_simul in a separate mod file. However, when I reach any recession periods, stoch_simul crashes with unsatisfactory Blanchard & Kahn - indeterminacy, even though the estimation went without a problem.

I am setting all estimated parameters in a loop via set_param_value. So I have 2 questions:

  1. Is there a better way to load the estimated results in a bulk?
  2. What could cause the indeterminacy in B&K conditions? Am I missing something? Perhaps some steady-state information from the estimation?

Thank you very much for any insight!

2

Without seeing the specific example, it is impossible to tell. Something must be going wrong.

3

I’ll try being more specific and provide the code.

This is what I use to loop over the stoch_simul with every recursive estimation of each parameter:

for i = 1:96  %for 96 recursive estimations
    if first_time           
        dynare model_loss noclearall;
        first_time = 0;
        result_var(i).ygap = oo_.var(1,1);
        result_var(i).pi = oo_.var(2,2);
        result_var(i).r = oo_.var(3,3)/4;
        result_var(i).ugap = oo_.var(4,4);
        result_var(i).t = t{i};        
    else
        set_param_value('cpibar',p_cpibar(i));
        set_param_value('crhointr',p_crhointr(i)); 
        set_param_value('cry',p_cry(i)); 
        set_param_value('crdy',p_crdy(i)); 
        set_param_value('crpi',p_crpi(i));
        set_param_value('cphi', p_cphi(i));
        set_param_value('ch', p_ch(i));
        set_param_value('comega', p_comega(i));
        set_param_value('cv', p_cv(i));
        set_param_value('cthetap', p_cthetap(i));
        set_param_value('cthetaw', p_cthetaw(i));
        set_param_value('cgammap', p_cgammap(i));
        set_param_value('cgammaw', p_cgammaw(i));
        set_param_value('cpsip', p_cpsip(i));
        set_param_value('cpsiw', p_cpsiw(i));
        set_param_value('calpha', p_calpha(i));
        set_param_value('cthetae', p_cthetae(i));
        set_param_value('cbetabar', p_cbetabar(i));
        set_param_value('cebar', p_cebar(i));
        set_param_value('ctaubar', p_ctaubar(i));
        set_param_value('crhoa', p_crhoa(i));
        set_param_value('crhob', p_crhob(i));
        set_param_value('crhog', p_crhog(i));
        set_param_value('crhoga', p_crhoga(i));
        set_param_value('crhoq', p_crhoq(i));
        set_param_value('crhor', p_crhor(i));
        set_param_value('crhop', p_crhop(i));
        set_param_value('cmup', p_cmup(i));
        set_param_value('crhow', p_crhow(i));
        set_param_value('crhos', p_crhos(i));
        set_param_value('cmuw', p_cmuw(i));
        set_param_value('cpsi', p_cpsi(i));
        [info, oo_] = stoch_simul(M_, options_, oo_, var_list_);
        if info
          disp(['Computation fails for t = ' t{i}]);
        else
        result_var(i).ygap = oo_.var(1,1);
        result_var(i).pi = oo_.var(2,2);
        result_var(i).r = oo_.var(3,3)/4;
        result_var(i).ugap = oo_.var(4,4);
        end
    end        
end
for i = 1:96
    result_var(i).t = t{i};
end

With this being the mod file:
model_loss.mod (10.7 KB)

The stochastic simulation fails in some of the recession periods (2009 and 2019), I doubt that is a coincidence. I set parameter values of all estimated parameters, perhaps I am supposed to add something for each recursion?

I tried adding a steady_state_model block at the end, but that sadly did not help.

(edit: formatting)

4

Can you please also provide the parameter file to run the code.

5

Of course, I forgot about that, sorry.

This is the file:
irf_params_all_10y.mat (26.4 KB)

I load it simply by running:

load('irf_params_all_10y.mat')

6

Please upload a zip file with code I can run. The mat file does not contain the parameters you are trying to set.

7

Here it is professor. Hopefully the error is replicable.

Loss_functions.zip (30.8 KB)

8

You need

        options_.qz_criterium=1-1e-6;

in the second part of the loop. The problem is the different treatment of near unit roots in estimation and stoch_simul.

9

That solves the issue. Thank you, professor!