stepan
/
dynare
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
dynare/matlab/swz/sz_prd.m

554 lines
21 KiB
Matlab
Raw Blame History

function sz_prd(options_)
%==========================================================================
%== Directory structure
%==========================================================================
%generation of mhm file
generateMHM_input(options_);
swz_root = strrep(which('swz_sbvar'),'/swz_sbvar.m','');
% path for C executables
%c_path='./c-executables';
c_path = [swz_root '/bin'];
% path for Markov specification
m_spec_path=[swz_root '/switching_specification'];
% path for MHM specification
mhm_spec_path=[swz_root '/mhm_specification'];
%==========================================================================
%== Output control
%==========================================================================
% tag for output files %create an option
%options_.ms.output_file_tag='test_2v';
% 1 to create init_<options_.ms.output_file_tag>.dat, 0 otherwise
%options_.ms.create_initialization_file = 1; %0 originally
% 1 to perform estimation, 0 otherwise
%options_.ms.estimate_msmodel = 1;
% 1 to perform estimation, 0 otherwise
%options_.ms.compute_mdd = 1;
% 1 to compute probabilites, 0 otherwise
%options_.ms.compute_probabilities = 1;%1 in the original
% 1 to Prints draws of the posterior
%options_.ms.print_draws = 1;
%options_.ms.n_draws=1000;
%options_.ms.thinning_factor=1;
%==========================================================================
%== Markov Process Specification File
%==========================================================================
%options_.ms.markov_file = 'specification_2v2c.dat'; %create an option
markov_file = [options_.ms.markov_file '.dat'];
%==========================================================================
%== Markov Process Specification File
%==========================================================================
%options_.ms.mhm_file = 'MHM_input.dat';
mhm_file = [mhm_spec_path '/MHM_input.dat'];
%options_.ms.proposal_draws = 100000;
%==========================================================================
%== Var Specification
%==========================================================================
% Number of options_.ms.nlags
%options_.ms.nlags = 4;
% Var restriction function
%options_.ms.restriction_fname = 'ftd_upperchol3v'; %create an option
%==========================================================================
%== BVAR prior
%==========================================================================
%=== The following mu is effective only if indxPrior==1.
mu = zeros(6,1); % hyperparameters
% mu(1): overall tightness for A0 and Aplus
mu(1) = 1.0;
% mu(2): relative tightness for Aplus
mu(2) = 1.0;
% mu(3): relative tightness for the constant term
mu(3) = 0.1;
% mu(4): tightness on lag decay. (1.2 - 1.5 faster decay produces better
% inflation forecasts
mu(4) = 1.2;
% mu(5): weight on nvar sums of coeffs dummy observations (unit roots).
mu(5) = 1;
% mu(6): weight on single dummy initial observation including constant
% (cointegration, unit roots, and stationarity).
mu(6) = 1;
% Alpha on p. 66 for squared time-varying structural shock lambda.
galp = 1.0;
% Beta on p. 66 for squared time-varying structural shock lambda.
gbeta = 1.0;
% Case 3 (no state change across options_.ms.nlags (l) but allows all variables for a
% given lag to switch states). Normal prior variance for glamda
% (nvar-by-nvar for each state) for different variables in lagged D+. See
% p.71v.
gsig2_lmdm = 50^2;
%==========================================================================
%== Data
%==========================================================================
% Read in data to produce rectangular array named xdd. Each column is one
% data series.
%load ./data/artificial_data
xdd=options_.data;
% Information about timing of the data for consistancy checks
% quarters (4) or months (12)
%q_m = 4;
%options_.ms.freq = 4;
q_m = options_.ms.freq;
% beginning year in data set
%yrBin=1978;
%options_.ms.initial_year = 1959;
yrBin=options_.ms.initial_year;
% beginning quarter or month in data set
%qmBin=1;
%options_.ms.initial_subperiod = 1;
qmBin=options_.ms.initial_subperiod;
% final year in data set
%yrFin=2007;
%options_.ms.final_year = 2005;
yrFin=options_.ms.final_year;
% final month or quarter in data set
%qmFin=4;
%options_.ms.final_subperiod = 4;
qmFin=options_.ms.final_subperiod;
% first year to use in estimation
%yrStart=yrBin;
yrStart=options_.ms.initial_year;
% first quarter or month to use in estimation
%qmStart=qmBin;
qmStart=options_.ms.initial_subperiod;
% last year to use in estimation
%yrEnd=yrFin;
yrEnd=options_.ms.final_year;
% last quater or month to use in estimation
%qmEnd=qmFin;
qmEnd=options_.ms.final_subperiod;
% Log variables in xdd
logindx = [];
% Convert percent to decimal in xdd
pctindx = [];
% Select the variable to use and rearrange columns if desired
%vlist = [3 1 2];
%options_.ms.vlist = [1 2 3];
options_.ms.vlist = [1:size(options_.varobs,1)];
vlist1=options_.ms.vlist;
%==========================================================================
%== Linux or Windows system - differences in some naming conventions
%==========================================================================
use_linux = 1;
%==========================================================================
%== Random number seed for selecting starting point in constant parameter
%== optimization.
%==========================================================================
% Set to zero to set from clock
seednumber = 7910;
if seednumber
randn('state',seednumber);
rand('state',seednumber);
else
rand('state',fix(100*sum(clock)));
randn('state',1000000000*rand);
rand('state',1000000000*rand);
end
%==========================================================================
%==========================================================================
%==========================================================================
%== Beginning of code. Modify below at own risk.
%==========================================================================
% options that may at some point become user specified
%indxC0Pres = 0; % 1: cross-A0-and-A+ restrictions; 0: idfile_const is all we have
indxC0Pres =options_.ms.cross_restrictions;
% Example for indxOres==1: restrictions of the form P(t) = P(t-1).
%Rform = 0; % 1: contemporaneous recursive reduced form; 0: restricted (non-recursive) form
Rform =options_.ms.contemp_reduced_form;
% % % Pseudo = 0; % 1: Pseudo forecasts; 0: real time forecasts
%indxPrior = 1; % 1: Bayesian prior; 0: no prior
indxPrior =options_.ms.bayesian_prior;
%indxDummy = indxPrior; % 1: add dummy observations to the data; 0: no dummy added.
indxDummy = options_.ms.bayesian_prior;
%ndobs = 0; % No dummy observations for xtx, phi, fss, xdatae, etc. Dummy observations are used as an explicit prior in fn_rnrprior_covres_dobs.m.
ndobs =options_.ms.dummy_obs;
%if indxDummy
% ndobs=nvar+1; % number of dummy observations
%else
% ndobs=0; % no dummy observations
%end
%
hpmsmd = [0.0; 0.0];
indxmsmdeqn = [0; 0; 0; 0]; %This option disenable using this in fn_rnrprior_covres_dobs.m
nStates = -1;
%==========================================================================
%== Create initialization file
%==========================================================================
if options_.ms.create_initialization_file == 1
%======================================================================
%== Check and setup data
%======================================================================
% log data
xdd(:,logindx) = log(xdd(:,logindx));
% convert percentage to decimal
xdd(:,pctindx)=.01*xdd(:,pctindx);
if (q_m ~= 12) && (q_m ~= 4)
disp('Warning: data must be monthly or quarterly!')
return
end
% number of data points
nData=(yrFin-yrBin)*q_m + (qmFin-qmBin+1);
% number of data points in estimation sample
nSample=(yrEnd-yrStart)*q_m + (qmEnd-qmEnd+1);
% number of periods not used at beginning of data (non-negative number)
nStart=(yrStart-yrBin)*q_m + (qmStart-qmBin);
% number of periods not used at end of data (non-positive number)
nEnd=(yrEnd-yrFin)*q_m + (qmEnd-qmFin);
if (nEnd > 0) || (nStart < 0)
disp('Warning: desired estimation period not in data set!')
return
end
if (nSample <= 0)
disp('Warning: no data points in estimation period!')
return
end
% reorder variables and create estimation data set
xdgel=xdd(nStart+1:nData+nEnd,vlist1);
% bad data points
baddata = find(isnan(xdgel));
if ~isempty(baddata)
disp('Warning: some data for estimation period are unavailable.')
return
end
% set nvar and nexo
nvar=size(xdgel,2);
nexo=1;
% 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);
%======================================================================
%== Linear Restrictions
%======================================================================
if Rform
Ui=cell(nvar,1); Vi=cell(ncoef,1);
for kj=1:nvar
Ui{kj} = eye(nvar); Vi{kj} = eye(ncoef);
end
else
eval(['[Ui,Vi,n0,np,ixmC0Pres] = ' options_.ms.restriction_fname '(options_.ms.nlags,nvar,nexo,indxC0Pres);'])
if min(n0)==0
disp('A0: restrictions give no free parameters in one of equations')
return
elseif min(np)==0
disp('Aplus: Restrictions in give no free parameters in one of equations')
return
end
end
%======================================================================
%== Estimation
%======================================================================
if indxPrior
%*** 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,q_m,options_.ms.nlags,xdgel,mu,indxDummy,hpmsmd,indxmsmdeqn);
else
[Pi,H0multi,Hpmulti,H0invmulti,Hpinvmulti] = fn_rnrprior(nvar,q_m,options_.ms.nlags,xdgel,mu);
end
%*** Combines asymmetric prior with linear restrictions
[Ptld,H0invtld,Hpinvtld] = fn_rlrprior(Ui,Vi,Pi,H0multi,Hpmulti,nvar);
%*** Obtains the posterior matrices for estimation and inference
[Pmat,H0inv,Hpinv] = fn_rlrpostr(xtx,xty,yty,Ptld,H0invtld,Hpinvtld,Ui,Vi);
else
%*** Obtain the posterior matrices for estimation and inference
[Pmat,H0inv,Hpinv] = fn_dlrpostr(xtx,xty,yty,Ui,Vi);
end
if Rform
%*** Obtain the ML estimate
A0hatinv = chol(H0inv{1}/fss); % upper triangular but lower triangular choleski
A0hat=inv(A0hatinv);
Aphat = Pmat{1}*A0hat;
else
%*** Obtain the ML estimate
% load idenml
x = 10*rand(sum(n0),1);
H0 = eye(sum(n0));
crit = 1.0e-9;
nit = 10000;
%
tic
[fhat,xhat,grad,Hhat,itct,fcount,retcodehat] = csminwel('fn_a0freefun',x,H0,'fn_a0freegrad',crit,nit,Ui,nvar,n0,fss,H0inv);
endtime = toc
A0hat = fn_tran_b2a(xhat,Ui,nvar,n0);
xhat = fn_tran_a2b(A0hat,Ui,nvar,n0);
[Aphat,ghat] = fn_gfmean(xhat,Pmat,Vi,nvar,ncoef,n0,np);
if indxC0Pres
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 = Vi{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 = Vi{ixeq}'*Fhatur0P(:,ixeq);
Aphat(:,ixeq) = Vi{ixeq}*(gihat + HLV*(ci-Lit*gihat));
end
end
end
%======================================================================
%== Create matlab initialization file
%======================================================================
matlab_filename = ['matlab_',options_.ms.output_file_tag,'.prn'];
fidForC = fopen(matlab_filename,'w');
fprintf(fidForC,'\n%s\n','//== gxia: alpha parameter for gamma prior of xi ==//');
fprintf(fidForC,' %20.15f ', galp);
fprintf(fidForC, '\n\n');
fprintf(fidForC,'\n%s\n','//== gxib: beta parameter for gamma prior of xi ==//');
fprintf(fidForC,' %20.15f ', gbeta);
fprintf(fidForC, '\n\n');
fprintf(fidForC,'\n%s\n','//== glamdasig: sigma parameter for normal prior of lamda ==//');
fprintf(fidForC,' %20.15f ', sqrt(gsig2_lmdm));
fprintf(fidForC, '\n\n');
%=== lags, nvar, nStates, sample size (excluding options_.ms.nlags where, with dummyies, fss=nSample-options_.ms.nlags+ndobs).
fprintf(fidForC,'\n%s\n','//== lags, nvar, nStates, T ==//');
fprintf(fidForC,' %d %d %d %d\n\n\n',options_.ms.nlags, nvar, nStates, fss);
%=== A0hat nvar-by-nvar from the constant VAR.
fprintf(fidForC,'\n%s\n','//== A0hat: nvar-by-nvar ==//');
indxFloat = 1;
xM = A0hat;
nrows = nvar;
ncols = nvar;
fn_fprintmatrix(fidForC, xM, nrows, ncols, indxFloat)
%=== Aphat ncoef-by-nvar from the constant VAR.
%=== Each column of Aphat is in the order of [nvar variables for 1st lag, ..., nvar variables for last lag, constant term].
fprintf(fidForC,'\n%s\n','//== Aphat: ncoef(lags*nvar+1)-by-nvar ==//');
indxFloat = 1;
xM = Aphat;
nrows = ncoef;
ncols = nvar;
fn_fprintmatrix(fidForC, xM, nrows, ncols, indxFloat)
%=== n0const: nvar-by-1, whose ith element represents the number of free A0 parameters in ith equation for the case of constant parameters.
fprintf(fidForC,'\n%s\n','//== n0const: nvar-by-1 ==//');
indxFloat = 0;
xM = n0;
nrows = 1;
ncols = nvar;
fn_fprintmatrix(fidForC, xM', nrows, ncols, indxFloat)
%=== npconst: nvar-by-1, whose ith element represents the number of free A+ parameters in ith equation for the case of constant parameters.
fprintf(fidForC,'\n%s\n','//== npconst: nvar-by-1 ==//');
indxFloat = 0;
xM = np;
nrows = 1;
ncols = nvar;
fn_fprintmatrix(fidForC, xM', nrows, ncols, indxFloat)
%=== Uiconst: nvar-by-1 cell. In each cell, nvar-by-qi orthonormal basis for the null of the ith
% equation contemporaneous restriction matrix where qi is the number of free parameters.
% With this transformation, we have ai = Ui*bi or Ui'*ai = bi where ai is a vector
% of total original parameters and bi is a vector of free parameters. When no
% restrictions are imposed, we have Ui = I. There must be at least one free
% parameter left for the ith equation.
fprintf(fidForC,'\n%s\n','//== Uiconst: cell(nvar,1) and nvar-by-n0const(i) for the ith cell (equation) ==//');
for i_=1:nvar
fn_fprintmatrix(fidForC, Ui{i_}, nvar, n0(i_), 1);
end
%=== Viconst: nvar-by-1 cell. In each cell, k-by-ri orthonormal basis for the null of the ith
% equation lagged restriction matrix where k is a total of exogenous variables and
% ri is the number of free parameters. With this transformation, we have fi = Vi*gi
% or Vi'*fi = gi where fi is a vector of total original parameters and gi is a
% vector of free parameters. There must be at least one free parameter left for
% the ith equation.
fprintf(fidForC,'\n%s\n','//== Viconst: cell(nvar,1) and ncoef-by-n0const(i) for the ith cell (equation) ==//');
for i_=1:nvar
fn_fprintmatrix(fidForC, Vi{i_}, ncoef, np(i_), 1);
end
%=== H0barconstcell: cell(nvar,1) (equations) and n-by-n for each cell (equaiton).
%=== H0barconst: prior covariance matrix for each column of A0 under asymmetric prior (including SZ dummy obs.) with NO linear restrictions imposed yet.
fprintf(fidForC,'\n%s\n','//== H0barconstcell: cell(nvar,1) and n-by-n for the ith cell (equation) ==//');
for i_=1:nvar
fn_fprintmatrix(fidForC, H0multi(:,:,i_), nvar, nvar, 1);
end
%=== Hpbarconstcell: cell(nvar,1) (equations) and ncoef-by-ncoef for each cell (equaiton).
%=== Hpbarconst: prior covariance matrix for each column of A+ under asymmetric prior (including SZ dummy obs.) with NO linear restrictions imposed yet.
fprintf(fidForC,'\n%s\n','//== Hpbarconstcell: cell(nvar,1) and ncoef-by-ncoef for the ith cell (equation) ==//');
for i_=1:nvar
fn_fprintmatrix(fidForC, Hpmulti(:,:,i_), ncoef, ncoef, 1);
end
%=== phi: X; T-by-k; column: [nvar for 1st lag, ..., nvar for last lag, other exogenous terms, const term]
fprintf(fidForC,'\n%s\n','//== Xright -- X: T-by-ncoef ==//');
xM = phi;
nrows = fss;
ncols = ncoef;
for ki=1:nrows
for kj=1:ncols
fprintf(fidForC,' %20.15f ',xM((kj-1)*nrows+ki));
if (kj==ncols)
fprintf(fidForC,'\n');
end
end
if (ki==nrows)
fprintf(fidForC,'\n\n');
end
end
%=== y: Y: T-by-nvar where T=fss
fprintf(fidForC,'\n%s\n','//== Yleft -- Y: T-by-nvar ==//');
xM = y;
nrows = fss;
ncols = nvar;
for ki=1:nrows
for kj=1:ncols
fprintf(fidForC,' %20.15f ',xM((kj-1)*nrows+ki));
if (kj==ncols)
fprintf(fidForC,'\n');
end
end
if (ki==nrows)
fprintf(fidForC,'\n\n');
end
end
fclose(fidForC);
%======================================================================
%== Create C initialization filename
%======================================================================
if use_linux == 1
create_init_file=[c_path,'/sbvar_init_file ',matlab_filename,' ',m_spec_path,'/',markov_file,' ',options_.ms.output_file_tag];
system(create_init_file) %Run operating system command and return result
else
create_init_file=[c_path,'\sbvar_init.exe ',matlab_filename,' ',m_spec_path,'\',markov_file,' ',options_.ms.output_file_tag];
dos(create_init_file)
end
end
%==========================================================================
%== Perform estimation
%==========================================================================
if options_.ms.estimate_msmodel == 1
if use_linux == 1
perform_estimation=[c_path,'/sbvar_estimate -ft ',options_.ms.output_file_tag];
system(perform_estimation)
else
perform_estimation=[c_path,'\sbvar_estimate.exe -ft ',options_.ms.output_file_tag];
dos(perform_estimation)
end
end
%==========================================================================
%== Compute marginal data density
%==========================================================================
if options_.ms.compute_mdd == 1
if use_linux == 1
compute_mdd1=[c_path,'/sbvar_mhm_1 -ft ',options_.ms.output_file_tag,' -fi ',mhm_spec_path,'/',mhm_file];
system(compute_mdd1);
compute_mdd2=[c_path,'/sbvar_mhm_2 -ft ',options_.ms.output_file_tag,' -d ',int2str(options_.ms.proposal_draws),' -t 3'];
system(compute_mdd2);
else
compute_mdd1=[c_path,'\sbvar_mhm_1.exe -ft ',options_.ms.output_file_tag,' -fi ',mhm_spec_path,'/',mhm_file];
system(compute_mdd1);
compute_mdd2=[c_path,'\sbvar_mhm_2.exe -ft ',options_.ms.output_file_tag,' -d ',int2str(options_.ms.proposal_draws),' -t 3'];
system(compute_mdd2);
end
end
%==========================================================================
%== Compute posterior mode regime probabilities
%==========================================================================
if options_.ms.compute_probabilities == 1 %error registers here
if use_linux == 1
compute_prob=[c_path,'/sbvar_probabilities -ft ',options_.ms.output_file_tag];
system(compute_prob);
else
compute_prob=[c_path,'\sbvar_probabilities -ft ',options_.ms.output_file_tag];
system(compute_prob);
end
end
%==========================================================================
%== Print Draws
%==========================================================================
if options_.ms.print_draws == 1 %error here as well
if use_linux == 1
print_draws=[c_path,'/sbvar_draws -ft ',options_.ms.output_file_tag,' -i ',int2str(options_.ms.n_draws),' -t ',int2str(options_.ms.thinning_factor)];
system(print_draws);
else
print_draws=[c_path,'\sbvar_draws -ft ',options_.ms.output_file_tag,' -i ',int2str(options_.ms.n_draws),' -t ',int2str(options_.ms.thinning_factor)];
system(print_draws);
end
end
Reference in New Issue View Git Blame Copy Permalink