812 lines
37 KiB
Matlab
812 lines
37 KiB
Matlab
%function msstart2(options_)
|
||
% This .m file is called for point graphics or error bands and
|
||
% It starts for both data and Bayesian estimation (when IxEstimat==0,
|
||
% no estimation but only data analysis), which allows you to set
|
||
% (a) available data range,
|
||
% (b) sample range,
|
||
% (c) rearrangement of actual data such as mlog, qg, yg
|
||
% (d) conditions of shocks 'Cms',
|
||
% (c) conditions of variables 'nconstr',
|
||
% (e) soft conditions 'nbancon,'
|
||
% (f) produce point conditional forecast (at least conditions on variables).
|
||
%
|
||
% February 2004
|
||
|
||
% Copyright © 2011-2017 Dynare Team
|
||
%
|
||
% This file is part of Dynare.
|
||
%
|
||
% Dynare is free software: you can redistribute it and/or modify
|
||
% it under the terms of the GNU General Public License as published by
|
||
% the Free Software Foundation, either version 3 of the License, or
|
||
% (at your option) any later version.
|
||
%
|
||
% Dynare is distributed in the hope that it will be useful,
|
||
% but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||
% GNU General Public License for more details.
|
||
%
|
||
% You should have received a copy of the GNU General Public License
|
||
% along with Dynare. If not, see <https://www.gnu.org/licenses/>.
|
||
|
||
% ** ONLY UNDER UNIX SYSTEM
|
||
%path(path,'/usr2/f1taz14/mymatlab')
|
||
|
||
%addpath('C:\SoftWDisk\MATLAB6p5\toolbox\cstz')
|
||
|
||
|
||
msstart_setup
|
||
%===========================================
|
||
% Exordium II
|
||
%===========================================
|
||
%options_.ms.ncsk = 0; % conditional directly on shoocks. Unlike Cms, not on variables that back out shocks
|
||
%options_.ms.nstd = 6; % number of standard deviations to cover the range in which distributions are put to bins
|
||
%options_.ms.ninv = 1000; % the number of bins for grouping, say, impulse responses
|
||
%options_.ms.Indxcol = [1:nvar]; % a vector of random columns in which MC draws are made.
|
||
%
|
||
%options_.ms.indxparr = 1; % 1: parameters random; 0: no randomness in parameters
|
||
% Note, when 0, there is no effect from the values of options_.ms.IndxAp, options_.ms.Aband, etc.
|
||
%options_.ms.indxovr = 0; % 1: distributions for other variables of interest; 0: no distribution.
|
||
% Example: joint distribution of a(1) and a(2). Only for specific purposes
|
||
%options_.ms.Aband = 1; % 1: error bands with only A0 and A+ random.
|
||
%options_.ms.IndxAp = 1; % 1: generate draws of A+; 0: no such draws.
|
||
% Note: when options_.ms.IndxAp=0, there is no effect from the values of options_.ms.options_.ms.options_.ms.options_.ms.indximf, IndxFore,
|
||
% or options_.ms.apband.
|
||
%options_.ms.apband = 1; % 1: error bands for A+; 0: no error bands for A+.
|
||
%*** The following (impulse responses and forecasts) is used only if options_.ms.IndxAp=1
|
||
%options_.ms.indximf = 1; % 1: generate draws of impulse responses; 0: no such draws (thus no effect
|
||
% from options_.ms.imfband)
|
||
%options_.ms.imfband = 1; % 1: error bands for impulse responses; 0: no error bands
|
||
%options_.ms.indxfore = 0; % 1: generate draws of forecasts; 0: no such draws (thus no effect from options_.ms.foreband)
|
||
%options_.ms.foreband = 0; % 1: error bands for out-of-sample forecasts; 0: no error bands
|
||
%
|
||
%options_.ms.indxgforhat = 1; % 1: plot unconditoinal forecasts; 0: no such plot
|
||
rnum = nvar; % number of rows in the graph
|
||
cnum = 1; % number of columns in the graph
|
||
if rnum*cnum<nvar
|
||
warning('rnum*cnum must be at least as large as nvar')
|
||
disp('Hit any key to continue, or ctrl-c to abort')
|
||
pause
|
||
end
|
||
%options_.ms.indxgimfhat = 1; % 1: plot ML impulse responses; 0: no plot
|
||
%options_.ms.indxestima = 1; %1: ML estimation; 0: no estimation and data only
|
||
%
|
||
IndxNmlr = [1 0 0 0 0 0]; % imported by nmlzvar.m
|
||
% Index for which normalization rule to choose
|
||
% Only one of the elments in IndxNmlr can be non-zero
|
||
% IndxNmlr(1): ML A distance rule (supposed to be the best)
|
||
% IndxNmlr(2): ML Ahat distance rule (to approximate IndxNmlr(1))
|
||
% IndxNmlr(3): ML Euclidean distance rule (not invariant to scale)
|
||
% IndxNmlr(4): Positive diagonal rule
|
||
% IndxNmlr(5): Positive inv(A) diagonal rule (if ~IndxNmlr(5), no need for A0inu,
|
||
% so let A0inu=[])
|
||
% IndxNmlr(6): Assigned postive rule (such as off-diagonal elements). Added 1/3/00
|
||
|
||
|
||
%%%%----------------------------------------
|
||
% Hard conditions directly on variables
|
||
%
|
||
%options_.ms.indxgdls = 1; % 1: graph point forecast on variables; 0: disable
|
||
nconstr1=nfqm; % number of the 1st set of constraints
|
||
nconstr2=options_.forecast ; % number of the 2nd set of constraints
|
||
nconstr=nconstr1+nconstr2; % q: 4 years -- 4*12 months.
|
||
% When 0, no conditions directly on variables <<>>
|
||
nconstr=0 ; %6*nconstr1;
|
||
options_.ms.eq_ms = []; % location of MS equation; if [], all shocks
|
||
PorR = [4*ones(nconstr1,1);2*ones(nconstr1,1);3*ones(nconstr1,1)]; % the variable conditioned. 1: Pcm; 3: FFR; 4: CPI
|
||
PorR = [PorR;1*ones(nconstr1,1);5*ones(nconstr1,1);6*ones(nconstr1,1)];
|
||
%PorR = [3 5]; % the variable conditioned. 3: FFR; 5: CPI
|
||
%PorR = 3;
|
||
|
||
%%%%----------------------------------------
|
||
% Conditions directly on future shocks
|
||
%
|
||
%options_.ms.cms = 0 % 1: condition on ms shocks; 0: disable this and "fidcnderr.m" gives
|
||
% unconditional forecasts if nconstr = 0 as well; <<>>
|
||
%options_.ms.ncms = 0; % number of the stance of policy; 0 if no tightening or loosening
|
||
%options_.ms.eq_cms = 1; % location of MS shocks
|
||
options_.ms.tlindx = 1*ones(1,options_.ms.ncms); % 1-by-options_.ms.ncms vector; 1: tightening; 0: loosen
|
||
options_.ms.tlnumber = [0.5 0.5 0 0]; %94:4 % [2 2 1.5 1.5]; %79:9 %[1.5 1.5 1 1]; 90:9
|
||
% 1-by-options_.ms.ncms vector; cut-off point for MS shocks
|
||
TLmean = zeros(1,options_.ms.ncms);
|
||
% unconditional, i.e., 0 mean, for the final report in the paper
|
||
if options_.ms.cms
|
||
options_.ms.eq_ms = [];
|
||
% At least at this point, it makes no sense to have DLS type of options_.ms.eq_ms; 10/12/98
|
||
if all(isfinite(options_.ms.tlnumber))
|
||
for k=1:options_.ms.ncms
|
||
TLmean(k) = lcnmean(options_.ms.tlnumber(k),options_.ms.tlindx(k));
|
||
% shock mean magnitude. 1: tight; 0: loose
|
||
% Never used for any subsequent computation but
|
||
% simply used for the final report in the paper.
|
||
%options_.ms.tlnumber(k) = fzero('lcutoff',0,[],[],TLmean(k))
|
||
% get an idea about the cutoff point given TLmean instead
|
||
|
||
end
|
||
end
|
||
else
|
||
options_.ms.ncms = 0; % only for the use of the graph by msprobg.m
|
||
options_.ms.tlnumber = NaN*ones(1,options_.ms.ncms);
|
||
% -infinity, only for the use of the graph by msprobg.m
|
||
end
|
||
|
||
|
||
%%%%----------------------------------------
|
||
% Soft conditions on variables
|
||
%
|
||
%cnum = 0 % # of band condtions; when 0, disable this option
|
||
% Note (different from "fidencon") that each condition corres. to variable
|
||
%options_.ms.banact = 1; % 1: use infor on actual; 0: preset without infor on actual
|
||
if cnum
|
||
banindx = cell(cnum,1); % index for each variable or conditon
|
||
banstp = cell(cnum,1); % steps: annual in general
|
||
banvar = zeros(cnum,1); % varables: annual in general
|
||
banval = cell(cnum,1); % band value (each variable occupy a cell)
|
||
badval{1} = zeros(length(banstp{1}),2); % 2: lower or higher bound
|
||
|
||
banstp{1} = 1:4; % 3 or 4 years
|
||
banvar(1) = 3; % 3: FFR; 5: CPI
|
||
if ~options_.ms.banact
|
||
for i=1:length(banstp{1})
|
||
banval{1}(i,:) = [5.0 10.0];
|
||
end
|
||
end
|
||
end
|
||
%
|
||
pause(1)
|
||
skipline()
|
||
disp('For uncondtional forecasts, set nconstr = options_.ms.cms = cnum = 0')
|
||
pause(1)
|
||
%
|
||
%=================================================
|
||
%====== End of exordium ==========================
|
||
%=================================================
|
||
|
||
|
||
|
||
|
||
|
||
%(1)--------------------------------------
|
||
% Further data analysis
|
||
%(1)--------------------------------------
|
||
%
|
||
if (options_.ms.freq==12)
|
||
nStart=(yrStart-options_.ms.initial_year )*12+qmStart-options_.ms.initial_subperiod ; % positive number of months at the start
|
||
nEnd=(yrEnd-options_.ms.final_year )*12+qmEnd-options_.ms.final_subperiod ; % negative number of months towards end
|
||
elseif (options_.ms.freq==4)
|
||
nStart=(yrStart-options_.ms.initial_year )*4+qmStart-options_.ms.initial_subperiod ; % positive number of months at the start
|
||
nEnd=(yrEnd-options_.ms.final_year )*4+qmEnd-options_.ms.final_subperiod ; % negative number of months towards end
|
||
elseif (options_.ms.freq==1)
|
||
nStart=(yrStart-options_.ms.initial_year )*1+qmStart-options_.ms.initial_subperiod ; % positive number of months at the start
|
||
nEnd=(yrEnd-options_.ms.final_year )*1+qmEnd-options_.ms.final_subperiod ; % negative number of months towards end
|
||
else
|
||
error('Error: this code is only good for monthly/quarterly/yearly data!!!')
|
||
return
|
||
end
|
||
%
|
||
if nEnd>0 || nStart<0
|
||
disp('Warning: this particular sample consider is out of bounds of the data!!!')
|
||
return
|
||
end
|
||
%*** Note, both xdgel and xdata have the same start with the specific sample
|
||
xdgel=options_.data(nStart+1:nData+nEnd,options_.ms.vlist);
|
||
% gel: general options_.data within sample (nSample)
|
||
if ~(nSample==size(xdgel,1))
|
||
warning('The sample size (including options_.ms.nlags ) and data are incompatible')
|
||
disp('Check to make sure nSample and size(xdgel,1) are the same')
|
||
return
|
||
end
|
||
%
|
||
baddata = find(isnan(xdgel));
|
||
if ~isempty(baddata)
|
||
warning('Some data for this selected sample are actually unavailable.')
|
||
disp('Hit any key to continue, or ctrl-c to abort')
|
||
pause
|
||
end
|
||
%
|
||
if options_.ms.initial_subperiod ==1
|
||
yrB = options_.ms.initial_year ; qmB = options_.ms.initial_subperiod ;
|
||
else
|
||
yrB = options_.ms.initial_year +1; qmB = 1;
|
||
end
|
||
yrF = options_.ms.final_year ; qmF = options_.ms.final_subperiod ;
|
||
[Mdate,tmp] = fn_calyrqm(options_.ms.freq,[options_.ms.initial_year options_.ms.initial_subperiod ],[options_.ms.final_year options_.ms.final_subperiod ]);
|
||
xdatae=[Mdate options_.data(1:nData,options_.ms.vlist)];
|
||
% beyond sample into forecast horizon until the end of the data options_.ms.final_year :options_.ms.final_subperiod
|
||
% Note: may contain NaN data. So must be careful about its use
|
||
|
||
%=========== Obtain prior-period, period-to-last period, and annual growth rates
|
||
[yactyrge,yactyre,yactqmyge,yactqmge,yactqme] = fn_datana(xdatae,options_.ms.freq,options_.ms.log_var,options_.ms.percent_var,[yrB qmB],[yrF qmF]);
|
||
qdates = zeros(size(yactqmyge,1),1);
|
||
for ki=1:length(qdates)
|
||
qdates(ki) = yactqmyge(1,1) + (yactqmyge(1,2)+ki-2)/options_.ms.freq;
|
||
end
|
||
for ki=1:nvar
|
||
figure
|
||
plot(qdates, yactqmyge(:,2+ki)/100)
|
||
xlabel(options_.ms.varlist{ki})
|
||
end
|
||
save outactqmygdata.prn yactqmyge -ascii
|
||
|
||
|
||
|
||
%=========== Write the output on the screen or to a file in an organized way ==============
|
||
%disp([sprintf('%4.0f %2.0f %8.3f %8.3f %8.3f %8.3f %8.3f %8.3f %8.3f %8.3f\n',yactyrge')])
|
||
spstr1 = 'disp([sprintf(';
|
||
spstr2 = '%4.0f %2.0f';
|
||
yactyrget=yactyrge';
|
||
for ki=1:length(options_.ms.vlist)
|
||
if ki==length(options_.ms.vlist)
|
||
spstr2 = [spstr2 ' %8.3f\n'];
|
||
else
|
||
spstr2 = [spstr2 ' %8.3f'];
|
||
end
|
||
end
|
||
spstr = [spstr1 'spstr2' ', yactyrget)])'];
|
||
eval(spstr)
|
||
|
||
%
|
||
fid = fopen('outyrqm.prn','w');
|
||
%fprintf(fid,'%4.0f %2.0f %8.3f %8.3f %8.3f %8.3f %8.3f %8.3f %8.3f %8.3f\n',yactyrge');
|
||
fpstr1 = 'fprintf(fid,';
|
||
fpstr2 = '%4.0f %2.0f';
|
||
for ki=1:nvar
|
||
if ki==nvar
|
||
fpstr2 = [fpstr2 ' %8.3f\n'];
|
||
else
|
||
fpstr2 = [fpstr2 ' %8.3f'];
|
||
end
|
||
end
|
||
fpstr = [fpstr1 'fpstr2' ', yactyrget);'];
|
||
eval(fpstr)
|
||
fclose(fid);
|
||
|
||
|
||
|
||
if options_.ms.indxestima
|
||
%(2)----------------------------------------------------------------------------
|
||
% Estimation
|
||
% ML forecast and impulse responses
|
||
% Hard or soft conditions for conditional forecasts
|
||
%(2)----------------------------------------------------------------------------
|
||
%
|
||
%* Arranged data information, WITHOUT dummy obs when 0 after mu is used. See fn_rnrprior_covres_dobs.m for using the dummy
|
||
% observations as part of an explicit prior.
|
||
[xtx,xty,yty,fss,phi,y,ncoef,xr,Bh] = fn_dataxy(nvar,options_.ms.nlags ,xdgel,mu,0,nexo);
|
||
if qmStart+options_.ms.nlags -options_.ms.dummy_obs >0
|
||
qmStartEsti = rem(qmStart+options_.ms.nlags -options_.ms.dummy_obs ,options_.ms.freq); % dummy observations are included in the sample.
|
||
if (~qmStartEsti)
|
||
qmStartEsti = options_.ms.freq;
|
||
end
|
||
yrStartEsti = yrStart + floor((qmStart+options_.ms.nlags -options_.ms.dummy_obs )/(options_.ms.freq+0.01));
|
||
% + 0.01 (or any number < 1) is used so that qmStart+options_.ms.nlags -options_.ms.dummy_obs ==?*options_.ms.freq doesn't give us an extra year forward.
|
||
else
|
||
qmStartEsti = options_.ms.freq + rem(qmStart+options_.ms.nlags -options_.ms.dummy_obs ,options_.ms.freq); % dummy observations are included in the sample.
|
||
if (qmStart+options_.ms.nlags -options_.ms.dummy_obs ==0)
|
||
yrStartEsti = yrStart - 1; % one year back.
|
||
else
|
||
yrStartEsti = yrStart + floor((qmStart+options_.ms.nlags -options_.ms.dummy_obs )/(options_.ms.freq-0.01));
|
||
% - 0.01 (or any number < 1) is used so that qmStart+options_.ms.nlags -options_.ms.dummy_obs ==-?*options_.ms.freq give us an extra year back.
|
||
end
|
||
end
|
||
dateswd = fn_dataext([yrStartEsti qmStartEsti],[yrEnd qmEnd],xdatae(:,1:2)); % dates with dummies
|
||
phie = [dateswd phi];
|
||
ye = [dateswd y];
|
||
|
||
%* Obtain linear restrictions
|
||
[Uiconst,Viconst,n0,np,ixmC0Pres] = feval(options_.ms.restriction_fname,nvar,nexo,options_.ms );
|
||
if min(n0)==0
|
||
skipline()
|
||
warning('A0: restrictions in dlrprior.m give no free parameter in one of equations')
|
||
disp('Press ctrl-c to abort')
|
||
pause
|
||
elseif min(np)==0
|
||
skipline()
|
||
warning('Ap: Restrictions in dlrprior.m give no free parameter in one of equations')
|
||
disp('Press ctrl-c to abort')
|
||
pause
|
||
end
|
||
|
||
if options_.ms.contemp_reduced_form
|
||
Uiconst=cell(nvar,1); Viconst=cell(ncoef,1);
|
||
for kj=1:nvar
|
||
Uiconst{kj} = eye(nvar); Viconst{kj} = eye(ncoef);
|
||
end
|
||
end
|
||
|
||
if options_.ms.bayesian_prior
|
||
%*** Obtains asymmetric prior (with no linear restrictions) with dummy observations as part of an explicit prior (i.e,
|
||
% reflected in Hpmulti and Hpinvmulti). See Forecast II, pp.69a-69b for details.
|
||
if 1 % Liquidity effect prior on both MS and MD equations.
|
||
[Pi,H0multi,Hpmulti,H0invmulti,Hpinvmulti] = fn_rnrprior_covres_dobs(nvar,options_.ms.freq,options_.ms.nlags ,xdgel,mu,indxDummy,hpmsmd,indxmsmdeqn);
|
||
else
|
||
[Pi,H0multi,Hpmulti,H0invmulti,Hpinvmulti] = fn_rnrprior(nvar,options_.ms.freq,options_.ms.nlags ,xdgel,mu);
|
||
end
|
||
|
||
%*** Combines asymmetric prior with linear restrictions
|
||
[Ptld,H0invtld,Hpinvtld] = fn_rlrprior(Uiconst,Viconst,Pi,H0multi,Hpmulti,nvar);
|
||
|
||
%*** Obtains the posterior matrices for estimation and inference
|
||
[Pmat,H0inv,Hpinv] = fn_rlrpostr(xtx,xty,yty,Ptld,H0invtld,Hpinvtld,Uiconst,Viconst);
|
||
|
||
if options_.ms.contemp_reduced_form
|
||
%*** Obtain the ML estimate
|
||
A0hatinv = chol(H0inv{1}/fss); % upper triangular but lower triangular choleski
|
||
A0hat=inv(A0hatinv);
|
||
a0indx = find(A0hat);
|
||
else
|
||
%*** Obtain the ML estimate
|
||
% load idenml
|
||
x = 10*rand(sum(n0),1);
|
||
H0 = eye(sum(n0));
|
||
crit = 1.0e-9;
|
||
nit = 10000;
|
||
%
|
||
[fhat,xhat,grad,Hhat,itct,fcount,retcodehat] = ...
|
||
csminwel('fn_a0freefun',x,H0,'fn_a0freegrad',crit,nit,Uiconst,nvar,n0,fss,H0inv);
|
||
|
||
A0hat = fn_tran_b2a(xhat,Uiconst,nvar,n0)
|
||
A0hatinv = inv(A0hat);
|
||
fhat
|
||
xhat
|
||
grad
|
||
itct
|
||
fcount
|
||
retcodehat
|
||
save outm.mat xhat A0hat A0hatinv grad fhat itct itct fcount retcodehat
|
||
end
|
||
else
|
||
%*** Obtain the posterior matrices for estimation and inference
|
||
[Pmat,H0inv,Hpinv] = fn_dlrpostr(xtx,xty,yty,Uiconst,Viconst);
|
||
|
||
if options_.ms.contemp_reduced_form
|
||
%*** Obtain the ML estimate
|
||
A0hatinv = chol(H0inv{1}/fss); % upper triangular but lower triangular choleski
|
||
A0hat=inv(A0hatinv);
|
||
a0indx = find(A0hat);
|
||
else
|
||
%*** Obtain the ML estimate
|
||
% load idenml
|
||
x = 10*rand(sum(n0),1);
|
||
H0 = eye(sum(n0));
|
||
crit = 1.0e-9;
|
||
nit = 10000;
|
||
%
|
||
[fhat,xhat,grad,Hhat,itct,fcount,retcodehat] = ...
|
||
csminwel('fn_a0freefun',x,H0,'fn_a0freegrad',crit,nit,Uiconst,nvar,n0,fss,H0inv);
|
||
|
||
A0hat = fn_tran_b2a(xhat,Uiconst,nvar,n0)
|
||
A0hatinv = inv(A0hat);
|
||
fhat
|
||
xhat
|
||
grad
|
||
itct
|
||
fcount
|
||
retcodehat
|
||
save outm.mat xhat A0hat A0hatinv grad fhat itct itct fcount retcodehat
|
||
end
|
||
end
|
||
|
||
%**** impulse responses
|
||
swish = A0hatinv; % each column corresponds to an equation
|
||
if options_.ms.contemp_reduced_form
|
||
xhat = A0hat(a0indx);
|
||
Bhat=Pmat{1};
|
||
Fhat = Bhat*A0hat
|
||
ghat = NaN;
|
||
else
|
||
xhat = fn_tran_a2b(A0hat,Uiconst,nvar,n0);
|
||
[Fhat,ghat] = fn_gfmean(xhat,Pmat,Viconst,nvar,ncoef,n0,np);
|
||
if options_.ms.cross_restrictions
|
||
Fhatur0P = Fhat; % ur: unrestriced across A0 and A+
|
||
for ki = 1:size(ixmC0Pres,1) % loop through the number of equations in which
|
||
% cross-A0-A+ restrictions occur. See St. Louis Note p.5.
|
||
ixeq = ixmC0Pres{ki}(1,1); % index for the jth equation in consideration.
|
||
Lit = Viconst{ixeq}(ixmC0Pres{ki}(:,2),:); % transposed restriction matrix Li
|
||
% V_j(i,:) in f_j(i) = V_j(i,:)*g_j
|
||
ci = ixmC0Pres{ki}(:,4) .* A0hat(ixmC0Pres{ki}(:,3),ixeq);
|
||
% s * a_j(h) in the restriction f_j(i) = s * a_j(h).
|
||
LtH = Lit/Hpinv{ixeq};
|
||
HLV = LtH'/(LtH*Lit');
|
||
gihat = Viconst{ixeq}'*Fhatur0P(:,ixeq);
|
||
Fhat(:,ixeq) = Viconst{ixeq}*(gihat + HLV*(ci-Lit*gihat));
|
||
end
|
||
end
|
||
Fhat
|
||
Bhat = Fhat/A0hat; % ncoef-by-nvar reduced form lagged parameters.
|
||
end
|
||
nn = [nvar options_.ms.nlags imstp];
|
||
imfhat = fn_impulse(Bhat,swish,nn); % in the form that is congenial to RATS
|
||
imf3hat=reshape(imfhat,size(imfhat,1),nvar,nvar);
|
||
% imf3: row--steps, column--nvar responses, 3rd dimension--nvar shocks
|
||
imf3shat=permute(imf3hat,[1 3 2]);
|
||
% imf3s: permuted so that row--steps, column--nvar shocks,
|
||
% 3rd dimension--nvar responses
|
||
% Note: reshape(imf3s(1,:,:),nvar,nvar) = A0in (columns -- equations)
|
||
if options_.ms.indxgimfhat
|
||
figure
|
||
end
|
||
scaleout = fn_imcgraph(imfhat,nvar,imstp,xlab,ylab,options_.ms.indxgimfhat);
|
||
imfstd = max(abs(scaleout)'); % row: nvar (largest number); used for standard deviations
|
||
|
||
%
|
||
% %**** save stds. of both data and impulse responses in idfile1
|
||
% temp = [std(yactqmyge(:,3:end)); std(yactyrge(:,3:end)); imfstd]; %<<>>
|
||
% save idenyimstd.prn temp -ascii % export forecast and impulse response to the file "idenyimstd.prn", 3-by-nvar
|
||
% %
|
||
% %**** save stds. of both data and impulse responses in idfile1
|
||
% temp = [std(yactqmyge(:,3:end)); std(yactyrge(:,3:end)); imfstd]; %<<>>
|
||
% save idenyimstd.prn temp -ascii % export forecast and impulse response to the file "idenyimstd.prn", 3-by-nvar
|
||
% if options_.ms.indxparr
|
||
% idfile1='idenyimstd';
|
||
% end
|
||
|
||
%=====================================
|
||
% Now, out-of-sample forecasts. Note: Hm1t does not change with A0.
|
||
%=====================================
|
||
%
|
||
% * updating the last row of X (phi) with the current (last row of) y.
|
||
tcwx = nvar*options_.ms.nlags ; % total coefficeint without exogenous variables
|
||
phil = phi(size(phi,1),:);
|
||
phil(nvar+1:tcwx) = phil(1:tcwx-nvar);
|
||
phil(1:nvar) = y(end,:);
|
||
%*** exogenous variables excluding constant terms
|
||
if (nexo>1)
|
||
Xexoe = fn_dataext([yrEnd qmEnd],[yrEnd qmEnd],xdatae(:,[1:2 2+nvar+1:2+nvar+nexo-1]));
|
||
phil(1,tcwx+1:tcwx+nexo-1) = Xexoe(1,3:end);
|
||
end
|
||
%
|
||
%*** ML unconditional point forecast
|
||
nn = [nvar options_.ms.nlags nfqm];
|
||
if nexo<2
|
||
yforehat = fn_forecast(Bhat,phil,nn); % nfqm-by-nvar, in log
|
||
else
|
||
Xfexoe = fn_dataext(fdates(1,:),fdates(numel(fdates),:),xdatae(:,[1:2 2+nvar+1:2+nvar+nexo-1]));
|
||
%Xfexoe = fn_dataext(fdates(1,:),fdates(end,:),xdatae(:,[1:2 2+nvar+1:2+nvar+nexo-1]));
|
||
yforehat = fn_forecast(Bhat,phil,nn,nexo,Xfexoe(:,3:end)); % nfqm-by-nvar, in log
|
||
end
|
||
yforehate = [fdates yforehat];
|
||
%
|
||
yact1e = fn_dataext([yrEnd-nayr 1],[yrEnd qmEnd],xdatae(:,1:nvar+2));
|
||
if options_.ms.real_pseudo_forecast
|
||
%yact2e = fn_dataext([yrEnd-nayr 1],E2yrqm,xdatae);
|
||
yact2e = fn_dataext([yrEnd-nayr 1],[fdates(end,1) options_.ms.freq],xdatae(:,1:nvar+2));
|
||
else
|
||
yact2e=yact1e;
|
||
end
|
||
yafhate = [yact1e; yforehate]; % actual and forecast
|
||
%
|
||
%===== Converted to mg, qg, and calendar yg
|
||
%
|
||
[yafyrghate,yafyrhate,yafqmyghate] = fn_datana(yafhate,options_.ms.freq,options_.ms.log_var(1:nlogeno),options_.ms.percent_var(1:npereno));
|
||
% actual and forecast growth rates
|
||
[yact2yrge,yact2yre,yact2qmyge] = fn_datana(yact2e,options_.ms.freq,options_.ms.log_var(1:nlogeno),options_.ms.percent_var(1:npereno));
|
||
% only actual growth rates
|
||
yafyrghate
|
||
if options_.ms.indxgforhat
|
||
keyindx = 1:nvar;
|
||
conlab='unconditional';
|
||
|
||
figure
|
||
yafyrghate(:,3:end) = yafyrghate(:,3:end)/100;
|
||
yact2yrge(:,3:end) = yact2yrge(:,3:end)/100;
|
||
fn_foregraph(yafyrghate,yact2yrge,keyindx,rnum,cnum,options_.ms.freq,ylab,forelabel,conlab)
|
||
end
|
||
|
||
%-------------------------------------------------
|
||
% Setup for point conditional forecast
|
||
% ML Conditional Forecast
|
||
%-------------------------------------------------
|
||
%
|
||
%% See Zha's note "Forecast (1)" p. 5, RATS manual (some errors in RATS), etc.
|
||
%
|
||
%% Some notations: y(t+1) = y(t)B1 + e(t+1)inv(A0). e(t+1) is 1-by-n.
|
||
%% Let r(t+1)=e(t+1)inv(A0) + e(t+2)C + .... where inv(A0) is impulse
|
||
%% response at t=1, C at t=2, etc. The row of inv(A0) or C is
|
||
%% all responses to one shock.
|
||
%% Let r be q-by-1 (such as r(1) = r(t+1)
|
||
%% = y(t+1) (constrained) - y(t+1) (forecast)).
|
||
%% Use impulse responses to find out R (k-by-q) where k=nvar*nsteps
|
||
%% where nsteps the largest constrained step. The key of the program
|
||
%% is to creat R using impulse responses
|
||
%% Optimal solution for shock e where R'*e=r and e is k-by-1 is
|
||
%% e = R*inv(R'*R)*r.
|
||
%
|
||
|
||
if (nconstr > 0)
|
||
%*** initializing
|
||
stepcon=cell(nconstr,1); % initializing, value y conditioned
|
||
valuecon=zeros(nconstr,1); % initializing, value y conditioned
|
||
varcon=zeros(nconstr,1); % initializing, endogous variables conditioned
|
||
varcon(:)=PorR; % 1: Pcm; 3: FFR; 5: CPI
|
||
|
||
%
|
||
for i=1:nconstr
|
||
if i<=nconstr1
|
||
stepcon{i}=i; % FFR
|
||
elseif i<=2*nconstr1
|
||
stepcon{i}=i-nconstr1; % FFR
|
||
elseif i<=3*nconstr1
|
||
stepcon{i}=i-2*nconstr1; % FFR
|
||
elseif i<=4*nconstr1
|
||
stepcon{i}=i-3*nconstr1; % FFR
|
||
elseif i<=5*nconstr1
|
||
stepcon{i}=i-4*nconstr1; % FFR
|
||
elseif i<=6*nconstr1
|
||
stepcon{i}=i-5*nconstr1; % FFR
|
||
end
|
||
end
|
||
|
||
% for i=1:nconstr
|
||
% stepcon{i}=i; % FFR
|
||
% end
|
||
|
||
% bend=12;
|
||
% stepcon{1}=[1:bend]'; % average over
|
||
% stepcon{nconstr1+1}=[1:options_.ms.freq-qmSub]'; % average over the remaing months in 1st forecast year
|
||
% stepcon{nconstr1+2}=[options_.ms.freq-qmSub+1:options_.ms.freq-qmSub+12]'; % average over 12 months next year
|
||
% stepcon{nconstr1+3}=[options_.ms.freq-qmSub+13:options_.ms.freq-qmSub+24]'; % average over 12 months. 3rd year
|
||
% stepcon{nconstr1+4}=[options_.ms.freq-qmSub+25:options_.ms.freq-qmSub+36]'; % average over 12 months. 4th year
|
||
|
||
% %**** avearage condition over, say, options_.ms.freq periods
|
||
% if qmEnd==options_.ms.freq
|
||
% stepcon{1}=[1:options_.ms.freq]'; % average over the remaing periods in 1st forecast year
|
||
% else
|
||
% stepcon{1}=[1:options_.ms.freq-qmEnd]'; % average over the remaing periods in 1st forecast year
|
||
% end
|
||
% for kj=2:nconstr
|
||
% stepcon{kj}=[length(stepcon{kj-1})+1:length(stepcon{kj-1})+options_.ms.freq]'; % average over 12 months next year
|
||
% end
|
||
|
||
if options_.ms.real_pseudo_forecast
|
||
% %*** conditions in every period
|
||
% for i=1:nconstr
|
||
% valuecon(i) = yact(actup+i,varcon(i));
|
||
% %valuecon(i) = mean( yact(actup+1:actup+bend,varcon(i)) );
|
||
% %valuecon(i) = 0.060; % 95:01
|
||
% %valuecon(i) = (0.0475+0.055)/2; % 94:10
|
||
% end
|
||
|
||
% %*** average condtions over,say, options_.ms.freq periods.
|
||
% for i=nconstr1+1:nconstr1+nconstr2
|
||
% i=1;
|
||
% valuecon(nconstr1+i) = ( ( mean(ylast12Cal(:,varcon(nconstr1+i)),1) + ...
|
||
% log(1+yactCalyg(yAg-yFg+i,varcon(nconstr1+i))/100) )*options_.ms.freq - ...
|
||
% yCal_1(:,varcon(nconstr1+i)) ) ./ length(stepcon{nconstr1+i});
|
||
% % the same as unconditional "yactCalyg" 1st calendar year
|
||
% i=2;
|
||
% valuecon(nconstr1+i) = mean(ylast12Cal(:,varcon(nconstr1+i))) + ...
|
||
% log(1+yactCalyg(yAg-yFg+1,varcon(nconstr1+i))/100) ...
|
||
% + log(1+yactCalyg(yAg-yFg+i,varcon(nconstr1+i))/100);
|
||
% % the same as actual "yactCalgy" 2nd calendar year
|
||
% i=3;
|
||
% valuecon(nconstr1+i) = valuecon(nconstr1+i-1) + ...
|
||
% log(1+yactCalyg(yAg-yFg+i,varcon(nconstr1+i))/100);
|
||
% % the same as actual "yactCalgy" 3rd calendar year
|
||
% %i=4;
|
||
% %valuecon(nconstr1+i) = valuecon(nconstr1+i-1) + ...
|
||
% % log(1+yactCalyg(yAg-yFg+i,varcon(nconstr1+i))/100);
|
||
% % the same as actual "yactCalgy" 4th calendar year
|
||
% end
|
||
|
||
%*** conditions in every period
|
||
vpntM = fn_dataext(E1yrqm, E2yrqm,xdatae); % point value matrix with dates
|
||
% vaveM = fn_dataext([yrEnd+1 0],[yrEnd+options_.forecast 0],yact2yre); % average value matrix with dates
|
||
for i=1:nconstr
|
||
if i<=nconstr1
|
||
valuecon(i) = vpntM(i,2+varcon(i)); % 2: first 2 elements are dates
|
||
elseif i<=2*nconstr1
|
||
valuecon(i) = vpntM(i-nconstr1,2+varcon(i));
|
||
elseif i<=3*nconstr1
|
||
valuecon(i) = vpntM(i-2*nconstr1,2+varcon(i));
|
||
elseif i<=4*nconstr1
|
||
valuecon(i) = vpntM(i-3*nconstr1,2+varcon(i));
|
||
elseif i<=5*nconstr1
|
||
valuecon(i) = vpntM(i-4*nconstr1,2+varcon(i));
|
||
elseif i<=6*nconstr1
|
||
valuecon(i) = vpntM(i-5*nconstr1,2+varcon(i));
|
||
end
|
||
end
|
||
|
||
% %*** average condtions over,say, options_.ms.freq periods.
|
||
% if qmEnd==options_.ms.freq
|
||
% vaveM = fn_dataext([yrEnd+1 0],[yrEnd+options_.forecast 0],yact2yre); % average value matrix with dates
|
||
% valuecon(1) = vaveM(1,2+varcon(1)); % 2: first 2 elements are dates
|
||
% else
|
||
% vaveM = fn_dataext([yrEnd 0],[yrEnd+options_.forecast 0],yact2yre); % average value matrix with dates
|
||
% yactrem = fn_dataext([yrEnd qmEnd+1],[yrEnd options_.ms.freq],xdatae);
|
||
% valuecon(1) = sum(yactrem(:,2+varcon(1)),1)/length(stepcon{1});
|
||
% % 2: first 2 elements are dates
|
||
% end
|
||
% for kj=2:nconstr
|
||
% valuecon(kj) = vaveM(kj,2+varcon(kj)); % 2: first 2 elements are dates
|
||
% end
|
||
else
|
||
vpntM = dataext([yrEnd qmEnd+1],[yrEnd qmEnd+2],xdatae); % point value matrix with dates
|
||
for i=1:nconstr
|
||
if i<=nconstr1
|
||
valuecon(i) = vpntM(i,2+varcon(i)); % 2: first 2 elements are dates; Poil
|
||
elseif i<=2*nconstr1
|
||
valuecon(i) = vpntM(i-nconstr1,2+varcon(i)); % 2: first 2 elements are dates; M2
|
||
elseif i<=3*nconstr1
|
||
valuecon(i) = vpntM(i-2*nconstr1,2+varcon(i)); % 2: first 2 elements are dates; FFR
|
||
elseif i<=4*nconstr1
|
||
valuecon(i) = vpntM(i-3*nconstr1,2+varcon(i)); % 2: first 2 elements are dates; CPI
|
||
elseif i<=5*nconstr1
|
||
valuecon(i) = vpntM(i-4*nconstr1,2+varcon(i)); % 2: first 2 elements are dates; U
|
||
elseif i<=5*nconstr1+nconstr2
|
||
valuecon(i)=xdata(end,5)+(i-5*nconstr1)*log(1.001)/options_.ms.freq; %CPI
|
||
elseif i<=5*nconstr1+2*nconstr2
|
||
valuecon(i)=0.0725; %FFR
|
||
else
|
||
valuecon(i)=xdata(end,6)+(i-5*nconstr1-2*nconstr2)*0.01/nfqm; %U
|
||
end
|
||
end
|
||
%valuecon(i) = 0.060; % 95:01
|
||
end
|
||
else
|
||
valuecon = [];
|
||
stepcon = [];
|
||
varcon = [];
|
||
end
|
||
|
||
nstepsm = 0; % initializing, the maximum step in all constraints
|
||
for i=1:nconstr
|
||
nstepsm = max([nstepsm max(stepcon{i})]);
|
||
end
|
||
|
||
if cnum
|
||
if options_.ms.real_pseudo_forecast && options_.ms.banact
|
||
for i=1:length(banstp{1})
|
||
banval{1}(1:length(banstp{1}),1) = ...
|
||
yactCalyg(yAg-yFg+1:yAg-yFg+length(banstp{1}),banvar(1)) - 2;
|
||
banval{1}(1:length(banstp{1}),2) = ...
|
||
yactCalyg(yAg-yFg+1:yAg-yFg+length(banstp{1}),banvar(1)) + 2;
|
||
end
|
||
end
|
||
end
|
||
|
||
|
||
%===================================================
|
||
% ML conditional forecast
|
||
%===================================================
|
||
%/*
|
||
[ychat,Estr,rcon] = fn_fcstidcnd(valuecon,stepcon,varcon,nstepsm,...
|
||
nconstr,options_.ms.eq_ms,nvar,options_.ms.nlags ,phil,0,0,yforehat,imf3shat,A0hat,Bhat,...
|
||
nfqm,options_.ms.tlindx,options_.ms.tlnumber,options_.ms.ncms,options_.ms.eq_cms);
|
||
ychate = [fdates ychat];
|
||
yachate = [yact1e; ychate]; % actual and condtional forecast
|
||
%===== Converted to mg, qg, and calendar yg
|
||
[yacyrghate,yacyrhate,yacqmyghate] = fn_datana(yachate,options_.ms.freq,options_.ms.log_var(1:nlogeno),options_.ms.percent_var(1:npereno));
|
||
% actual and conditional forecast growth rates
|
||
if options_.ms.indxgdls && nconstr
|
||
keyindx = 1:nvar;
|
||
% conlab=['conditional on' ylab{PorR(1)}];
|
||
conlab='v-conditions';
|
||
|
||
figure
|
||
fn_foregraph(yafyrghate,yact2yrge,keyindx,rnum,cnum,options_.ms.freq,ylab,forelabel,conlab)
|
||
end
|
||
|
||
if options_.ms.ncsk
|
||
Estr = zeros(nfqm,nvar);
|
||
Estr(1:2,:) = [
|
||
-2.1838 -1.5779 0.53064 -0.099425 -0.69269 -1.0391
|
||
1.9407 3.3138 -0.10563 -0.55457 -0.68772 1.3534
|
||
];
|
||
Estr(3:6,3) = [0.5*ones(1,4)]'; % MD shocks
|
||
|
||
Estr(3:10,2) = [1.5 1.5 1.5*ones(1,6)]'; % MS shocks
|
||
|
||
%Estr(3:6,6) = 1*ones(4,1); % U shocks
|
||
%Estr(8:11,4) = 1*ones(4,1); % y shocks
|
||
|
||
%Estr(3:10,2) = [2.5 2.5 1.5*ones(1,6)]'; % MS shocks alone
|
||
|
||
nn = [nvar options_.ms.noptions_.ms.nlags nfqm];
|
||
ycEhat = forefixe(A0hat,Bhat,phil,nn,Estr);
|
||
ycEhate = [fdates ycEhat];
|
||
yacEhate = [yact1e; ycEhate]; % actual and condtional forecast
|
||
%===== Converted to mg, qg, and calendar yg
|
||
[yacEyrghate,yacEyrhate,yacEqmyghate] = datana(yacEhate,options_.ms.freq,options_.ms.log_var(1:nlogeno),options_.ms.percent_var(1:npereno));
|
||
% actual and conditional forecast growth rates
|
||
disp([sprintf('%4.0f %2.0f %8.3f %8.3f %8.3f %8.3f %8.3f %8.3f\n',yacEyrghate')])
|
||
|
||
if 1
|
||
keyindx = 1:nvar;
|
||
% conlab=['conditional on' ylab{PorR(1)}];
|
||
conlab='shock-conditions';
|
||
|
||
figure
|
||
gyrfore(yacEyrghate,yact2yrge,keyindx,rnum,cnum,ylab,forelabel,conlab)
|
||
end
|
||
end
|
||
|
||
%-----------------------------------------------------------
|
||
% Compute structural shocks for the whole sample period excluding dummy observations.
|
||
%-----------------------------------------------------------
|
||
ywod = y(options_.ms.dummy_obs +1:end,:); % without dummy observations
|
||
phiwod=phi(options_.ms.dummy_obs +1:end,:); % without dummy observations
|
||
eplhat=ywod*A0hat-phiwod*Fhat;
|
||
qmStartWod = mod(qmStart+options_.ms.nlags ,options_.ms.freq);
|
||
if (~qmStartWod)
|
||
qmStartWod = options_.ms.freq;
|
||
end
|
||
yrStartWod = yrStart + floor((qmStart+options_.ms.nlags -1)/options_.ms.freq);
|
||
dateswod = fn_dataext([yrStartWod qmStartWod],[yrEnd qmEnd],xdatae(:,1:2));
|
||
eplhate = [dateswod eplhat];
|
||
|
||
Aphat = Fhat;
|
||
end
|
||
|
||
%----------------------------------------
|
||
% Tests for LR, HQ, Akaike, Schwarz. The following gives a guidance.
|
||
% But the computation has to be done in a different M file by exporting fhat's
|
||
% from different idfile's.
|
||
%----------------------------------------
|
||
%
|
||
%if ~options_.ms.contemp_reduced_form
|
||
% SpHR=A0in'*A0in;
|
||
%end
|
||
%%
|
||
%if ~isnan(SpHR) && ~options_.ms.contemp_reduced_form
|
||
% warning(' ')
|
||
% disp('Make sure you run the program with options_.ms.contemp_reduced_form =1 first.')
|
||
% disp('Otherwise, the following test results such as Schwartz are incorrect.')
|
||
% disp('All other results such as A0ml and imfs, however, are correct.')
|
||
% disp('Press anykey to contintue or ctrl-c to stop now')
|
||
% pause
|
||
|
||
% load SpHUout
|
||
|
||
% logLHU=-fss*sum(log(diag(chol(SpHU)))) -0.5*fss*nvar % unrestricted logLH
|
||
|
||
% logLHR=-fhat % restricted logLH
|
||
% tra = reshape(SpHU,nvar*nvar,1)'*reshape(A0*A0',nvar*nvar,1);
|
||
% df=(nvar*(nvar+1)/2 - length(a0indx));
|
||
% S=2*(logLHU-logLHR);
|
||
% SC = (nvar*(nvar+1)/2 - length(a0indx)) * log(fss);
|
||
% disp(['T -- effective sample size: ' num2str(fss)])
|
||
% disp(['Trace in the overidentified posterior: ' num2str(tra)])
|
||
% disp(['Chi2 term -- 2*(logLHU-logLHR): ' num2str(S)])
|
||
% disp(['Degrees of freedom: ' num2str(df)])
|
||
% disp(['SC -- df*log(T): ' num2str(SC)])
|
||
% disp(['Akaike -- 2*df: ' num2str(2*df)])
|
||
% disp(['Classical Asymptotic Prob at chi2 term: ' num2str(cdf('chi2',S,df))])
|
||
|
||
% %*** The following is the eigenanalysis in the difference between
|
||
% %*** unrestricted (U) and restricted (R)
|
||
% norm(A0'*SpHU*A0-diag(diag(ones(6))))/6;
|
||
% norm(SpHU-A0in'*A0in)/6;
|
||
|
||
% corU = corr(SpHU);
|
||
% corR = corr(SpHR);
|
||
|
||
% [vU,dU]=eigsort(SpHU,1);
|
||
% [vR,dR]=eigsort(SpHR,1);
|
||
|
||
% [log(diag(dU)) log(diag(dR)) log(diag(dU))-log(diag(dR))];
|
||
|
||
% sum(log(diag(dU)));
|
||
% sum(log(diag(dR)));
|
||
%else
|
||
% disp('To run SC test, turn options_.ms.contemp_reduced_form =1 first and then turn options_.ms.contemp_reduced_form =0')
|
||
%end
|
||
|
||
|
||
%***** Simply regression
|
||
%X=[phi(:,3) y(:,2)-phi(:,2) y(:,1)-phi(:,7) ones(fss,1)];
|
||
%<25> Y=y(:,3);
|
||
%<25> b=regress(Y,X)
|
||
|
||
%=== Computes the roots for the whole system.
|
||
rootsinv = fn_varoots(Bhat,nvar,options_.ms.nlags )
|
||
abs(rootsinv)
|
||
|
||
|
||
bhat =xhat;
|
||
n0const=n0; % For constant parameter models.
|
||
n0const=n0; % For constant parameter models.
|
||
npconst=np; % For constant parameter models.
|
||
save outdata_a0dp_const.mat A0hat bhat Aphat n0const
|