update unstable

bvartools/OrderingIndexes.m

@@ -30,6 +30,7 @@
     if check == 1
         %assignin('base', [newstrng 'index_' varordering{ii} ], ii);
         String = varordering{ii};
+        String = replace(String,'-','');
         iString = String(find(~isspace(String)));
         assignin('base', [newstrng 'i' iString ], ii);
     else

bvartools/bvar_.m

@@ -723,9 +723,9 @@
 else % pooled units
     yy = []; XX = [];
     for nunit = 1 : nunits
-        [yy,XX] = YXB_(y(idx, :, nunit),lags,[nx timetrend]);
-        XX = [XX; XX];
-        yy = [yy; yy];
+        [yy_p,XX_p] = YXB_(y(idx, :, nunit),lags,[nx timetrend]);
+        XX = [XX; XX_p];
+        yy = [yy; yy_p];
     end
 end
 
@@ -1517,17 +1517,17 @@
 % inference and IRFs
 BVAR.Phi_draws    = Phi_draws;          % draws from the autoregressive part
 BVAR.Sigma_draws  = Sigma_draws;        % draws from the covarance matrix
-BVAR.alpha_draws  = Phi_draws;          % Older name: draws from the autoregressive part
-BVAR.sigma_draws  = Sigma_draws;        % Older name: draws from the covarance matrix
+% BVAR.alpha_draws  = Phi_draws;          % Older name: draws from the autoregressive part
+% BVAR.sigma_draws  = Sigma_draws;        % Older name: draws from the covarance matrix
 BVAR.Sigma_lower_chol_draw = Sigma_lower_chol_draw;
 BVAR.ir_draws     = ir_draws;           % draws from the IRF with cholesky
 BVAR.irlr_draws   = irlr_draws;         % draws from the IRF with Long Run
 BVAR.Qlr_draws    = Qlr_draws;          % Long Run Rotation matrix
 BVAR.lags         = lags;               % lags
 BVAR.N            = ny;                 % number of variables
 BVAR.e_draws      = e_draws;            % residuals
-BVAR.e            = e_draws;            % backward compatible with earlier versions
-BVAR.fe_draws     = fe_draws;            % backward compatible with earlier versions
+% BVAR.e            = e_draws;            % backward compatible with earlier versions
+BVAR.fe_draws     = fe_draws;           % insample forecast errors
 
 BVAR.posterior    = posterior;
 BVAR.prior        = prior;             % priors used

bvartools/max_fevd.m

@@ -5,18 +5,28 @@
 if nargin < 6
     Kappa = 1000;
 end
+if length(i) ~= length(j)
+    error('each shock must have one target')
+end
 
 N    = size(Sigma,1);
-crit = nan(Kappa,1);
+crit = nan(Kappa,length(j));
 Q    = nan(N,N,Kappa);
 for k = 1 : Kappa
     Q(:,:,k) = generateQ(N);                % generate an orthonormal matrix
     FEVD     = fevd(h,Phi,Sigma,Q(:,:,k));  % compute the FEVD
     % Calculate the contribution of shock j, to variable i forecast error volatility, at horizon h
-    crit(k,1) = FEVD(i, j);
+    for j1 = 1 : length(j)
+        crit(k,j1) = FEVD(i(j1), j(j1));
+    end
 end
 % Pick the maximum
-[~,index] = max(crit);
-Qbar      = Q(:,:,index); 
+Qbar = nan(N,N,length(j));
+for j1 = 1 : length(j)
+    [~,index] = max(crit(:,j1));
+    tmp_notindx = setdiff([1:N] , j(j1));
+    Omega = Q(:,:,index);
+    Qbar(:,:,j1) = Omega(:,[j(j1) tmp_notindx]);
+end
 
 end

bvartools/u2nu.m

@@ -0,0 +1,11 @@
+function [nu] = u2nu(e_draws,Sigma_draws,Omega_draws)
+
+nu=zeros(size(e_draws));
+for dd = 1 : size(e_draws,3)
+    errors     = e_draws(:,:,dd);
+    A          = chol( Sigma_draws(:,:,dd),'lower');
+    nu(:,:,dd) = errors * inv( Omega_draws(:,:,dd)' * A'); %#ok<MINV>
+
+    % nu(:,:,dd) = errors / (  Omega_draws(:,:,dd)' * A'); % structural innovations
+end
+

examples/BVAR tutorial/example_2_minn.m

@@ -101,3 +101,25 @@
 
 %% ESSENCE: OPTIMALLY CHOSEN  AND  CHERRY  PICKED  CHOSEN  
 %% HYPER-PARAMETER VALUES MAKE  LITTLE  DIFFERENCE  FOR  SHAPE  OF  IRFS.
+
+%% 3.4 Use a presample as prior
+clear options;
+% run a VAR on presample data
+presample = 50; % 8 years of presample
+lags      = 6;
+bvar1     = bvar_(y(1:presample,:),lags);
+
+
+% use the VAR estimates to set the priors for the LP
+options.priors.name        = 'Conjugate';
+% posterior mean of the VAR AR coeff and constant
+options.priors.Phi.mean    = mean(bvar1.Phi_draws,3);
+% average variance of the AR coeff and constant
+options.priors.Phi.cov     = diag(mean(var(bvar1.Phi_draws,0,3),2));
+% posterior mean of the Covariance of the VAR residuals 
+options.priors.Sigma.scale = mean(bvar1.Sigma_draws,3);
+options.priors.Sigma.df    = size(bvar1.Phi_draws,1)-2;
+options.K                  = 1000;
+
+bvar2 = bvar_(y(presample+1:end,:),lags,options);
+