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

added test for too large standard deviation in beta distribution; 

cleaned up error messages related to priors; set error trap for
initilial likelihood Nan or complex
time-shift
Michel Juillard 2012-06-13 15:47:01 +02:00
parent a39a0b3b67
commit 3ff832ddcc
3 changed files with 19 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
matlab/dynare_estimation_init.m
View File

@ -335,7 +335,7 @@ nvx = estim_params_.nvx;
ncx = estim_params_.ncx; ncx = estim_params_.ncx;
nvn = estim_params_.nvn; nvn = estim_params_.nvn;
ncn = estim_params_.ncn; ncn = estim_params_.ncn;
if ~isempty(estim_params_.param_vals) if estim_params_.np
M.params(estim_params_.param_vals(:,1)) = xparam1(nvx+ncx+nvn+ncn+1:end); M.params(estim_params_.param_vals(:,1)) = xparam1(nvx+ncx+nvn+ncn+1:end);
end; end;
oo_.steady_state = evaluate_steady_state(oo_.steady_state,M,options_,oo_,steadystate_check_flag); oo_.steady_state = evaluate_steady_state(oo_.steady_state,M,options_,oo_,steadystate_check_flag);

5
matlab/initial_estimation_checks.m
View File

@ -51,6 +51,11 @@ end
% error('Options mode_compute=5 is not compatible with non linear filters or Dsge-VAR models!') % error('Options mode_compute=5 is not compatible with non linear filters or Dsge-VAR models!')
% end % end
% end % end
if isnan(fval)
error('The initial value of the likelihood is NaN')
elseif imag(fval)
error('The initial value of the likelihood is complex')
end
if info(1) > 0 if info(1) > 0
disp('Error in computing likelihood for initial parameter values') disp('Error in computing likelihood for initial parameter values')

37
matlab/set_prior.m
View File

@ -94,11 +94,7 @@ if nvn
bayestopt_.p3 = [ bayestopt_.p3; estim_params_.var_endo(:,8)]; bayestopt_.p3 = [ bayestopt_.p3; estim_params_.var_endo(:,8)];
bayestopt_.p4 = [ bayestopt_.p4; estim_params_.var_endo(:,9)]; bayestopt_.p4 = [ bayestopt_.p4; estim_params_.var_endo(:,9)];
bayestopt_.jscale = [ bayestopt_.jscale; estim_params_.var_endo(:,10)]; bayestopt_.jscale = [ bayestopt_.jscale; estim_params_.var_endo(:,10)];
if isempty(bayestopt_.name) bayestopt_.name = [ bayestopt_.name; cellstr(options_.varobs(estim_params_.var_endo(:,1),:))];
bayestopt_.name = cellstr(char(options_.varobs(estim_params_.var_endo(:,1),:)));
else
bayestopt_.name = cellstr(char(char(bayestopt_.name), options_.varobs(estim_params_.var_endo(:,1),:)));
end
end end
if ncx if ncx
xparam1 = [xparam1; estim_params_.corrx(:,3)]; xparam1 = [xparam1; estim_params_.corrx(:,3)];
@ -110,13 +106,9 @@ if ncx
bayestopt_.p3 = [ bayestopt_.p3; estim_params_.corrx(:,9)]; bayestopt_.p3 = [ bayestopt_.p3; estim_params_.corrx(:,9)];
bayestopt_.p4 = [ bayestopt_.p4; estim_params_.corrx(:,10)]; bayestopt_.p4 = [ bayestopt_.p4; estim_params_.corrx(:,10)];
bayestopt_.jscale = [ bayestopt_.jscale; estim_params_.corrx(:,11)]; bayestopt_.jscale = [ bayestopt_.jscale; estim_params_.corrx(:,11)];
if isempty(bayestopt_.name) bayestopt_.name = [bayestopt_.name; cellstr(['corr ' ...
bayestopt_.name = cellstr(char(char(strcat(cellstr(M_.exo_names(estim_params_.corrx(:,1),:)), ... deblank(M_.exo_names(estim_params_.corrx(:,1),:)) ...
',' , cellstr(M_.exo_names(estim_params_.corrx(:,2),:)))))); ', ' , deblank(M_.exo_names(estim_params_.corrx(:,2),:))])];
else
bayestopt_.name = cellstr(char(char(bayestopt_.name), char(strcat(cellstr(M_.exo_names(estim_params_.corrx(:,1),:)), ...
',' , cellstr(M_.exo_names(estim_params_.corrx(:,2),:))))));
end
end end
if ncn if ncn
if isequal(M_.H,0) if isequal(M_.H,0)
@ -132,13 +124,9 @@ if ncn
bayestopt_.p3 = [ bayestopt_.p3; estim_params_.corrn(:,9)]; bayestopt_.p3 = [ bayestopt_.p3; estim_params_.corrn(:,9)];
bayestopt_.p4 = [ bayestopt_.p4; estim_params_.corrn(:,10)]; bayestopt_.p4 = [ bayestopt_.p4; estim_params_.corrn(:,10)];
bayestopt_.jscale = [ bayestopt_.jscale; estim_params_.corrn(:,11)]; bayestopt_.jscale = [ bayestopt_.jscale; estim_params_.corrn(:,11)];
if isempty(bayestopt_.name) bayestopt_.name = [bayestiopt_.name; cellstr(['corr ' ...
bayestopt_.name = cellstr(char(char(strcat(cellstr(M_.endo_names(estim_params_.corrn(:,1),:)),... deblank(M_.exo_names(estim_params_.corrn(:,1),:)) ...
',' , cellstr(M_.endo_names(estim_params_.corrn(:,2),:)))))); ', ' , deblank(M_.exo_names(estim_params_.corrn(:,2),:))])];
else
bayestopt_.name = cellstr(char(char(bayestopt_.name), char(strcat(cellstr(M_.endo_names(estim_params_.corrn(:,1),:)),...
',' , cellstr(M_.endo_names(estim_params_.corrn(:,2),:))))));
end
end end
if np if np
xparam1 = [xparam1; estim_params_.param_vals(:,2)]; xparam1 = [xparam1; estim_params_.param_vals(:,2)];
@ -150,11 +138,7 @@ if np
bayestopt_.p3 = [ bayestopt_.p3; estim_params_.param_vals(:,8)]; bayestopt_.p3 = [ bayestopt_.p3; estim_params_.param_vals(:,8)];
bayestopt_.p4 = [ bayestopt_.p4; estim_params_.param_vals(:,9)]; bayestopt_.p4 = [ bayestopt_.p4; estim_params_.param_vals(:,9)];
bayestopt_.jscale = [ bayestopt_.jscale; estim_params_.param_vals(:,10)]; bayestopt_.jscale = [ bayestopt_.jscale; estim_params_.param_vals(:,10)];
if isempty(bayestopt_.name) bayestopt_.name = [bayestopt_.name; cellstr(M_.param_names(estim_params_.param_vals(:,1),:))];
bayestopt_.name = cellstr(char(M_.param_names(estim_params_.param_vals(:,1),:)));
else
bayestopt_.name = cellstr(char(char(bayestopt_.name),M_.param_names(estim_params_.param_vals(:,1),:)));
end
end end
bayestopt_.ub = ub; bayestopt_.ub = ub;
@ -175,6 +159,11 @@ for i=1:length(k)
end end
mu = (bayestopt_.p1(k(i))-bayestopt_.p3(k(i)))/(bayestopt_.p4(k(i))-bayestopt_.p3(k(i))); mu = (bayestopt_.p1(k(i))-bayestopt_.p3(k(i)))/(bayestopt_.p4(k(i))-bayestopt_.p3(k(i)));
stdd = bayestopt_.p2(k(i))/(bayestopt_.p4(k(i))-bayestopt_.p3(k(i))); stdd = bayestopt_.p2(k(i))/(bayestopt_.p4(k(i))-bayestopt_.p3(k(i)));
if stdd^2 > (1-mu)*mu
error(sprintf(['Error in prior for %s: in a beta distribution with ' ...
'mean %f, the standard error can''t be larger than' ...
' %f.'], bayestopt_.name{k(i)},mu,sqrt((1-mu)*mu)))
end
bayestopt_.p6(k(i)) = (1-mu)*mu^2/stdd^2 - mu ; bayestopt_.p6(k(i)) = (1-mu)*mu^2/stdd^2 - mu ;
bayestopt_.p7(k(i)) = bayestopt_.p6(k(i))*(1/mu-1) ; bayestopt_.p7(k(i)) = bayestopt_.p6(k(i))*(1/mu-1) ;
m = compute_prior_mode([ bayestopt_.p6(k(i)) , bayestopt_.p7(k(i)) , bayestopt_.p3(k(i)) , bayestopt_.p4(k(i)) ],1); m = compute_prior_mode([ bayestopt_.p6(k(i)) , bayestopt_.p7(k(i)) , bayestopt_.p3(k(i)) , bayestopt_.p4(k(i)) ],1);