stepan
/
dynare
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix variable sorting in with ML without selected_variables_only, which was erroneously in decision rule order 

Also fixes wrong output when selected_variables_only was used
time-shift
Johannes Pfeifer 2016-04-11 00:06:39 +02:00
parent 1bf684442c
commit df20a394fc
2 changed files with 87 additions and 31 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
matlab/convert_dyn_45_to_44.m
View File

@ -1,19 +1,19 @@
function oo_ = convert_dyn_45_to_44(M_, options_, oo_)
%function oo_ = convert_dyn_45_to_44(M_, options_, oo_
function oo_ = convert_dyn_45_to_44(M_, options_, oo_,bayestopt_)
%function oo_ = convert_dyn_45_to_44(M_, options_, oo_,bayestopt_)
% Converts oo_ from 4.5 to 4.4
%
% INPUTS
% M_ [struct] dynare model struct
% options_ [struct] dynare options struct
% oo_ [struct] dynare output struct
%
% bayestopt_ [struct] structure storing information about priors
% OUTPUTS
% oo_ [struct] dynare output struct
%
% SPECIAL REQUIREMENTS
% none
% Copyright (C) 2015 Dynare Team
% Copyright (C) 2015-16 Dynare Team
%
% This file is part of Dynare.
%
@ -143,6 +143,31 @@ if isfield(oo_,'FilteredVariables')
end
end
%% resort fields that are in declaration order to decision rule order (previous undocumented behavior)
if ~isempty(options_.nk) && options_.nk ~= 0
if ~((any(bayestopt_.pshape > 0) && options_.mh_replic) || (any(bayestopt_.pshape> 0) && options_.load_mh_file)) %no Bayesian estimation
positions_in_decision_order=oo_.dr.inv_order_var(bayestopt_.smoother_var_list(bayestopt_.smoother_saved_var_list));
if ~(options_.selected_variables_only && ~(options_.forecast > 0)) %happens only when selected_variables_only is not used
oo_.FilteredVariablesKStepAhead(:,positions_in_decision_order,:)=oo_.FilteredVariablesKStepAhead;
if ~isempty(PK) %get K-step ahead variances
oo_.FilteredVariablesKStepAheadVariances(:,positions_in_decision_order,positions_in_decision_order,:)=oo_.FilteredVariablesKStepAheadVariances;
end
if ~isempty(decomp)
oo_.FilteredVariablesShockDecomposition(:,positions_in_decision_order,:,:)=oo_.FilteredVariablesShockDecomposition;
end
else
fprintf('\nconvert_dyn_45_to_44:: Due to a bug in Dynare 4.4.3 with the selected_variables_only option, the previous behavior\n')
fprintf('convert_dyn_45_to_44:: cannot be restored for FilteredVariablesKStepAhead, FilteredVariablesKStepAheadVariances, and\n')
fprintf('convert_dyn_45_to_44:: FilteredVariablesShockDecomposition\n')
end
end
end
if options_.filter_covariance
oo_.Smoother.Variance(oo_.dr.inv_order_var,oo_.dr.inv_order_var,:)=oo_.Smoother.Variance;
end
%% set old field posterior_std and remove new field posterior_std_at_mode
if isfield(oo_,'posterior_std_at_mode')
oo_.posterior_std=oo_.posterior_std_at_mode;

85
matlab/write_smoother_results.m
View File

@ -111,55 +111,86 @@ if options_.filter_covariance
end
%get indices of smoothed variables
i_endo_in_dr = bayestopt_.smoother_saved_var_list;
if ~isempty(options_.nk) && options_.nk ~= 0 && (~((any(bayestopt_.pshape > 0) && options_.mh_replic) || (any(bayestopt_.pshape> 0) && options_.load_mh_file)))
i_endo_in_bayestopt_smoother_varlist = bayestopt_.smoother_saved_var_list;
i_endo_in_dr_matrices=bayestopt_.smoother_var_list(i_endo_in_bayestopt_smoother_varlist);
if ~isempty(options_.nk) && options_.nk ~= 0
%write deviations from steady state, add constant for observables later
oo_.FilteredVariablesKStepAhead = aK(options_.filter_step_ahead,i_endo_in_dr,:);
oo_.FilteredVariablesKStepAhead = aK(options_.filter_step_ahead,i_endo_in_dr_matrices,:);
if ~isempty(PK) %get K-step ahead variances
oo_.FilteredVariablesKStepAheadVariances = ...
PK(options_.filter_step_ahead,i_endo_in_dr,i_endo_in_dr,:);
PK(options_.filter_step_ahead,i_endo_in_dr_matrices,i_endo_in_dr_matrices,:);
end
if ~isempty(decomp) %get decomposition
oo_.FilteredVariablesShockDecomposition = ...
decomp(options_.filter_step_ahead,i_endo_in_dr,:,:);
decomp(options_.filter_step_ahead,i_endo_in_dr_matrices,:,:);
end
end
for dr_index=bayestopt_.smoother_saved_var_list'
i1 = oo_.dr.order_var(bayestopt_.smoother_var_list(dr_index)); %get indices of smoothed variables in name vector
for i_endo_in_bayestopt_smoother_varlist=bayestopt_.smoother_saved_var_list'
i_endo_in_dr=bayestopt_.smoother_var_list(i_endo_in_bayestopt_smoother_varlist);
i_endo_declaration_order = oo_.dr.order_var(i_endo_in_dr); %get indices of smoothed variables in name vector
%% Compute constant
if options_.loglinear == 1 %logged steady state must be used
constant_current_variable=repmat(log(ys(i1)),gend,1);
constant_current_variable=repmat(log(ys(i_endo_declaration_order)),gend,1);
elseif options_.loglinear == 0 %unlogged steady state must be used
constant_current_variable=repmat((ys(i1)),gend,1);
constant_current_variable=repmat((ys(i_endo_declaration_order)),gend,1);
end
oo_.SmoothedVariables.(deblank(M_.endo_names(i1,:)))=atT(dr_index,:)'+constant_current_variable;
oo_.SmoothedVariables.(deblank(M_.endo_names(i_endo_declaration_order,:)))=atT(i_endo_in_dr,:)'+constant_current_variable;
if ~isempty(options_.nk) && options_.nk > 0 && ~((any(bayestopt_.pshape > 0) && options_.mh_replic) || (any(bayestopt_.pshape> 0) && options_.load_mh_file))
oo_.FilteredVariables.(deblank(M_.endo_names(i1,:)))=squeeze(aK(1,dr_index,2:end-(options_.nk-1)));
oo_.FilteredVariables.(deblank(M_.endo_names(i_endo_declaration_order,:)))=squeeze(aK(1,i_endo_in_dr,2:end-(options_.nk-1)));
end
oo_.UpdatedVariables.(deblank(M_.endo_names(i1,:)))=updated_variables(dr_index,:)'+constant_current_variable;
oo_.UpdatedVariables.(deblank(M_.endo_names(i_endo_declaration_order,:)))=updated_variables(i_endo_in_dr,:)'+constant_current_variable;
end
%% Add trend and constant for observed variables
for pos_iter=1:length(bayestopt_.mf)
oo_.Smoother.Constant.(deblank(M_.endo_names(bayestopt_.mfys(pos_iter),:)))=constant_part(pos_iter,:);
oo_.SmoothedVariables.(deblank(M_.endo_names(bayestopt_.mfys(pos_iter),:)))=yf(pos_iter,:)';
if ~isempty(options_.nk) && options_.nk > 0 && ~((any(bayestopt_.pshape > 0) && options_.mh_replic) || (any(bayestopt_.pshape> 0) && options_.load_mh_file))
%filtered variable E_t(y_t+1) requires to shift trend by 1 period
oo_.FilteredVariables.(deblank(M_.endo_names(bayestopt_.mfys(pos_iter),:)))=...
squeeze(aK(1,bayestopt_.mf(pos_iter),2:end-(options_.nk-1)))...
+trend_constant_observables_filtered.filter_ahead_1(pos_iter,:)';
for filter_iter=1:length(options_.filter_step_ahead)
filter_step=options_.filter_step_ahead(filter_iter);
oo_.FilteredVariablesKStepAhead(filter_iter,bayestopt_.mf(pos_iter),1+filter_step:end-(max(options_.filter_step_ahead)-filter_step)) = ...
squeeze(aK(filter_step,bayestopt_.mf(pos_iter),1+filter_step:end-(max(options_.filter_step_ahead)-filter_step)))...
+trend_constant_observables_filtered.(['filter_ahead_' num2str(filter_step)])(pos_iter,:)';
if ismember(bayestopt_.mf(pos_iter),bayestopt_.smoother_var_list(bayestopt_.smoother_saved_var_list))
oo_.SmoothedVariables.(deblank(M_.endo_names(bayestopt_.mfys(pos_iter),:)))=yf(pos_iter,:)';
if ~isempty(options_.nk) && options_.nk > 0
%filtered variable E_t(y_t+1) requires to shift trend by 1 period
oo_.FilteredVariables.(deblank(M_.endo_names(bayestopt_.mfys(pos_iter),:)))=...
squeeze(aK(1,bayestopt_.mf(pos_iter),2:end-(options_.nk-1)))...
+trend_constant_observables_filtered.filter_ahead_1(pos_iter,:)';
for filter_iter=1:length(options_.filter_step_ahead)
filter_step=options_.filter_step_ahead(filter_iter);
oo_.FilteredVariablesKStepAhead(filter_iter,find(i_endo_in_dr_matrices==bayestopt_.mf(pos_iter)),1+filter_step:end-(max(options_.filter_step_ahead)-filter_step)) = ...
squeeze(aK(filter_step,bayestopt_.mf(pos_iter),1+filter_step:end-(max(options_.filter_step_ahead)-filter_step)))...
+trend_constant_observables_filtered.(['filter_ahead_' num2str(filter_step)])(pos_iter,:)';
end
end
%updated variables are E_t(y_t) so no trend shift is required
oo_.UpdatedVariables.(deblank(M_.endo_names(bayestopt_.mfys(pos_iter),:)))=...
updated_variables(bayestopt_.mf(pos_iter),:)'+trend_constant_observables(pos_iter,:)';
end
end
%% resort fields that are in decision rule order to declaration order
if ~isempty(options_.nk) && options_.nk ~= 0
positions_in_declaration_order=oo_.dr.order_var(bayestopt_.smoother_var_list(bayestopt_.smoother_saved_var_list));
if ~(options_.selected_variables_only && ~(options_.forecast > 0)) %happens only when selected_variables_only is not used
oo_.FilteredVariablesKStepAhead(:,positions_in_declaration_order,:)=oo_.FilteredVariablesKStepAhead;
if ~isempty(PK) %get K-step ahead variances
oo_.FilteredVariablesKStepAheadVariances(:,positions_in_declaration_order,positions_in_declaration_order,:)=oo_.FilteredVariablesKStepAheadVariances;
end
if ~isempty(decomp)
oo_.FilteredVariablesShockDecomposition(:,positions_in_declaration_order,:,:)=oo_.FilteredVariablesShockDecomposition;
end
else
positions_in_declaration_order=oo_.dr.order_var(bayestopt_.smoother_var_list(bayestopt_.smoother_saved_var_list));
[junk,sorted_index_declaration_order]=sort(positions_in_declaration_order);
oo_.FilteredVariablesKStepAhead(:,sorted_index_declaration_order,:)=oo_.FilteredVariablesKStepAhead;
if ~isempty(PK) %get K-step ahead variances
oo_.FilteredVariablesKStepAheadVariances(:,sorted_index_declaration_order,sorted_index_declaration_order,:)=oo_.FilteredVariablesKStepAheadVariances;
end
if ~isempty(decomp)
oo_.FilteredVariablesShockDecomposition(:,sorted_index_declaration_order,:,:)=oo_.FilteredVariablesShockDecomposition;
end
end
%updated variables are E_t(y_t) so no trend shift is required
oo_.UpdatedVariables.(deblank(M_.endo_names(bayestopt_.mfys(pos_iter),:)))=...
updated_variables(bayestopt_.mf(pos_iter),:)'+trend_constant_observables(pos_iter,:)';
end
if options_.filter_covariance
oo_.Smoother.Variance(oo_.dr.order_var,oo_.dr.order_var,:)=oo_.Smoother.Variance;
end
%% get smoothed shocks