1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352
|
function [A0hin,Hm1t,fss,ndobs,phi,y,nvar,ncoef,SpH,SpHsc,xxhpc,a0indx,na0p,...
idmat0] = szbvar(idfile,q_m,lags,nSample,nhp,xdgel)
% Estimating the Bayesian VAR of Sims and Zha:
% function [A0hin,Hm1t,fss,ndobs,phi,y,nvar,ncoef,SpH,SpHsc,xxhpc,a0indx,na0p,...
% idmat0] = szbvar(idfile,q_m,lags,nSample,nhp,xdgel)
%
% idfile: Identification filename such as "iden4" (with the extension ".prn").
% q_m: quarter or month
% lags: the maximum length of lag
% nSample: the sample size (including lags, of course)
% nhp: number of haperparameters
% imstp: number of impusle response steps
% xdgel: the general matrix of the original data (no manipulation involved)
% ------
% A0hin: chol(SpH) -- inverse of A0h, upper triangular, each column corresponds to an equation
% Hm1t: reduced form B
% fss: in-sample-size (for forecasting). Including dummy observations
% ndobs: number of dummy observations, ndobs=nvar+1
% phi: X in the form of y = X*B+U. Row: nSmaple-lags+ndobs. Column: ncoef
% y: y in the form y = X*B+U. Including the dummy observations too,
% T-lags+ndobs-by-nvar.
% nvar: number of variables
% ncoef: number of coefficients in *each* equation. RHS coefficients only, nvar*lags+1
% SpH: divided by T, the final S in the Waggoner-Zha exponential part of p(A0|y)
% SpHsc: upper triagular in the lower triangular Choleski of SpH*fss, for Wishart
% xxhpc: chol(X'X+inv(H_p_tilde)): upper triangular but its transpose
% is lower triagular Choleski
% a0indx: the index number for free parameters in A0. Column meaning equation
% na0p: number of free parameters in A0
% idmat0: identification matrix for A0; only 1's and 0's. Column meaning equation.
%
% Quick Revisions: May 2003. See H1p_1 on lines 301-304.
%
%
% Copyright (C) 1997-2012 Christopher A. Sims and Tao Zha
%
% This 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.
%
% It 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.
%
% If you did not received a copy of the GNU General Public License
% with this software, see <http://www.gnu.org/licenses/>.
%
%
%@@@ Prepared for Bayesian VAR of Sims and Zha
%
%* Load identification and obtain idmat0
eval(['load ' idfile '.prn -ascii']);
eval(['idmat0=' idfile ';']);
idmat0=idmat0'; % so that each column corresponds to an equation
a0indx=find(idmat0); % column meaning equation
na0p = length(a0indx); % number of free parameters in A0
nfp = na0p + nhp; % total number of free parameter
[nvar,neqn]=size(idmat0);
%$$$ available computations
%
%@@@ original
% total number of the sample under study, including lags, etc. -- original sample size
sb = lags+1; % original beginning without dummies
sl = nSample; % original last period without dummies
ssp = nSample - lags; % original number of observations
%
ndobs=nvar+1; % number of dummy observations
ncoef = nvar*lags+1; % number of coefficients in *each* equation, RHS coefficients only.
%*** initializing for global variables
%%GlbAph=zeros(nvar,ncoef);
Aplus=zeros(ncoef,nvar);
SpH = ones(nvar,nvar);
%* flp: forecast last period (with dummies);
%* fbp: forecast beginning period (with dummies).
flp = sl+ndobs;
fbp = ndobs+lags+1; % <<>> true begining by skipping dummies
fss = flp-lags; % forecast sample size (with dummies).
% ** hyperparameters
mu = zeros(nhp,1);
mu(1) = 0.57;
mu(2) = 0.13;
mu(3) = 0.1;
mu(4) = 1;
mu(5) = 5;
mu(6) = 5;
%
% mu(1): overall tightness and also for A0;
% mu(2): relative tightness for A+;
% mu(3): relative tightness for the constant term;
% mu(4): tightness on lag decay;
% mu(5): weight on nvar sums of coeffs dummy observations (unit roots);
% mu(6): weight on single dummy initial observation including constant (cointegration, unit
% roots, and stationarity).
% ** monthly lag decay in order to match quarterly decay: a*exp(bl) where
% ** l is the monthly lag. Suppose quarterly decay is 1/x where x=1,2,3,4.
% ** Let the decay of l1 match that of x1 (say, beginning) and the decay
% ** of l2 match that of x2 (say, end), we can solve for a and b which are
% ** b = (log_x1-log_x2)/(l1-l2), and a = x1*exp(-b*l1).
if q_m==12
l1 = 1; % 1st month == 1st quarter
xx1 = 1; % 1st quarter
l2 = lags; % last month
xx2 = 1/((ceil(lags/3))^mu(4)); % last quarter
%xx2 = 1/6; % last quarter
% 3rd quarter: i.e., we intend to let decay of the 6th month match
% that of the 3rd quarter, so that the 6th month decays a little
% faster than the second quarter which is 1/2.
if lags==1
b = 0;
else
b = (log(xx1)-log(xx2))/(l1-l2);
end
a = xx1*exp(-b*l1);
end
%
% ** now run the VAR with
%
% **** nvar prior dummy observations with the sum of coefficients
% ** construct X for Y = X*B + U where phi = X: (T-lags)*k, Y: (T-lags)*nvar
% ** columns: k = # of [nvar for 1st lag, ..., nvar for last lag, const]
% ** Now, T=T+ndobs -- added with "ndobs" dummy observations
%
phi = zeros(fss,ncoef);
const = ones(fss,1);
const(1:nvar) = zeros(nvar,1);
phi(:,ncoef) = const; % the first nvar periods: no or zero constant!
%
xdgelint = mean(xdgel(1:lags,:),1);
% mean of the first lags initial conditions
for k=1:nvar
for m=1:lags
phi(ndobs,nvar*(m-1)+k) = xdgelint(k);
phi(k,nvar*(m-1)+k) = xdgelint(k);
% <<>> multiply hyperparameter later
end
end
for k=1:lags
phi(ndobs+1:fss,nvar*(k-1)+1:nvar*k) = xdgel(sb-k:sl-k,:);
% row: T-lags; column: [nvar for 1st lag, ..., nvar for last lag, const]
% Thus, # of columns is nvar*lags+1 = ncoef.
end
%
% ** Y with "ndobs" dummies added
y = zeros(fss,nvar);
for k=1:nvar
y(ndobs,k) = xdgelint(k);
y(k,k) = xdgelint(k);
% <<>> place hyperparameter later
end
y(ndobs+1:fss,:) = xdgel(sb:sl,:);
%
% *** specify the prior for each equation separately, SZ method,
% ***
% *** obtaining the residuals for the univariate processes:
% *** a total of "nvar" equations
% *** obtain the posterior peak of A0 which is Sims iden1 (Sims 1986)
% ** get the residuals from univariate regressions.
%
sgh = zeros(nvar,1); % square root
sgsh = sgh; % square
nn = [1 lags nSample];
yu = xdgel;
C = ones(nSample,1);
for k=1:nvar
[Bk,ek,junk1,junk2,junk3,junk4] = sye([yu(:,k) C],nn);
clear Bk junk1 junk2 junk3 junk4;
sgsh(k) = ek'*ek/ssp;
%% sgh(k) = ek'*ek/(fss-7); to match "sqrt(ess)" in RATS univariate regression
sgh(k) = sqrt(sgsh(k));
end
% ** prior variance for alpha0, same for all equations!!!
al0b = zeros(nvar*neqn,0); % prior mean for all A0
sg0bid = zeros(nvar,1); % Sigma0_bar diagonal only
for j=1:nvar
sg0bid(j) = 1/sgsh(j); % sgsh = sigmai^2
end
% ** prior variance for alpha_plus, same for all equations
sgpbid = zeros(ncoef,1); % Sigma_plus_bar, diagonal
for i = 1:lags
if (q_m==12)
lagdecay = a*exp(b*i);
end
%
for j = 1:nvar
if (q_m==12)
% exponential decay to match quarterly decay
sgpbid((i-1)*nvar+j) = lagdecay^2/sgsh(j);
%%sgpbid((i-1)*nvar+j) = (1/i)^2/sgsh(j);
elseif (q_m==4)
sgpbid((i-1)*nvar+j) = (1/i^mu(4))^2/sgsh(j);
else
warning('Incompatibility with lags, check the possible errors!!!')
return
end
end
end
%%A0bd = sqrt(sg0bid);
% commented out by T.A. Zha, 10/3/96, to avoid double-counting of scaling and the problem
% of potential multiple peaks.
A0bd = zeros(nvar,1);
A0b = sparse(1:nvar,1:nvar,A0bd,nvar,nvar);
A0b = A0b'; % making a column in A0b correspond to an equation
%
%=================================================
% Computing the (prior) covariance matrix for the posterior of A0, no data yet
%=================================================
%
%
% ** weight prior dummy observations
%phi(1:nvar,:) = (mu(5)^2/mu(1)^2)*phi(1:nvar,:);
%y(1:nvar,:) = (mu(5)^2/mu(1)^2)*y(1:nvar,:);
%phi(ndobs,:) = mu
% modified by CAS 8/6/96. The dummy obs weights are naturally in std units, not var units.
%
phi(1:nvar,:) = mu(5)*phi(1:nvar,:); % standard Sims and Zha prior
y(1:nvar,:) = mu(5)*y(1:nvar,:); % standard Sims and Zha prior
%----- The following prior designed for GLZ
%phi(1,:) = 1.00*mu(5)*phi(1,:);
%phi(2:nvar,:) = 1.02*mu(7)*phi(2:nvar,:);
%y(1,:) = mu(5)*y(1,:);
%y(2:nvar,:) = mu(7)*y(2:nvar,:);
%----------------------------
%
phi(nvar+1,:) = mu(6)*phi(nvar+1,:);
y(nvar+1,:) = mu(6)*y(nvar+1,:);
% ** set up the conditional prior variance sg0bi and sgpbi.
sg0bida = mu(1)^2*sg0bid;
sgpbida = mu(1)^2*mu(2)^2*sgpbid;
sgpbida(ncoef) = mu(1)^2*mu(3)^2;
sgppbd = sgpbida(nvar+1:ncoef); % corresponding to A++, in a Sims-Zha paper
%%sgppbdi = 1./sgppbd;
%sgppb = diag(sgppbd);
Hptd = zeros(ncoef);
Hptdi=Hptd;
%%Hptd(nvar+1:ncoef,nvar+1:ncoef)=diag(sgppbd);
%%Hptdi(nvar+1:ncoef,nvar+1:ncoef)=diag(sgppbdi);
Hptd(ncoef,ncoef)=sgppbd(ncoef-nvar);
Hptdi(ncoef,ncoef)=1/sgppbd(ncoef-nvar);
% condtional on A0i, H_plus_tilde
sg0bd = sg0bida;
sg0bdi = 1./sg0bd;
sg1bd = sgpbida(1:nvar);
sg1bdi = 1./sg1bd;
if lags>1
sgpp_cbd = sgppbd(1:ncoef-nvar-1);
sgpp_cbdi = 1./sgpp_cbd; % _cbd: c -- constant, b -- barred (without), d -- diagonal
end
% ** set up the unconditional prior variance on A0i and A+i
%%XX = zeros(nvar,strm);
%%XX(stri,:) = eye(strm); % XX in Gelman, etel, on p479
%
% * final conditional prior variance on A+i give that on A0i, and
% * unconditional variance on A0+
H0td = diag(sg0bd); % unconditional
% H_~: td: tilde for ~
% ** inverse and chol decomp
H0tdi = diag(sg0bdi);
%
Hptd(1:nvar,1:nvar)=diag(sg1bd);
Hptdi(1:nvar,1:nvar)=diag(sg1bdi);
if lags>1
Hptd(nvar+1:ncoef-1,nvar+1:ncoef-1)=diag(sgpp_cbd);
Hptdi(nvar+1:ncoef-1,nvar+1:ncoef-1)=diag(sgpp_cbdi);
% condtional on A0i, H_plus_tilde
end
%=================================================
% Computing the final covariance matrix for the posterior of A0, with the data
%=================================================
%
% ** some common terms
[u d v]=svd(phi,0); %trial
% xtx = phi'*phi; %trial
vd=v.*(ones(size(v,2),1)*diag(d)'); %trial
xtx=vd*vd';
% M = eye(fss) - phi*(xtx\phi'); % not used except in line below %trial
yu = y'*u; %trial
cxyxx=yu*yu'; %trial
yty=y'*y;
ymy=yty-cxyxx; %trial
%ymy = y'*M*y;
%cxy = phi'*y; %trial
cxy = vd*yu'; %trial
cyx = cxy';
%cxpy = xtx\cxy; %not used further except in line below
%cxyxx = cxy'*cxpy;
% ** estimate A+_hat conditional on A0, ncoef*nvar, but not using full prior
%GlbAph = cxpy*A0h;
% common terms
xxhp = xtx+Hptdi;
%Lxxhpc = chol(inv(xxhp))'; % lower triangular
xxhpc = chol(xxhp); % upper triangular but its transpose is lower triagular Choleski
%A0hbH = A0hb'*H0tdi;
%====== The following two lines seem incorrect. The third line is supposed to correc the mistake. May 2003.
% H1p_1 = zeros(ncoef,nvar);
% H1p_1(1:nvar,:) = Hptdi(1:nvar,1:nvar);
H1p_1 = Hptdi(:,1:nvar);
%%Hm = (cyx+H1p_1')*(xxhp\(cxy+H1p_1));
Hm1 = (cyx+H1p_1')/xxhp;
Hm1t = Hm1';
%Hm2 = cxy+H1p_1;
Hm = Hm1*(cxy+H1p_1);
%%alpMpyh = A0h(stri,i)'*XX'*(yty+Hptdi(1:nvar,1:nvar)-Hm)*XX;
Hss = yty+Hptdi(1:nvar,1:nvar)-Hm;
SpH = (H0tdi+Hss)/fss; % the final S in the exponential part of p(A0|y)
% divided by nobs in order to make Choleski decomp without optimization
% or in the form conformable to Waggoner and Zha
% ***
% *** Form the inverse of covarinace to draw from t- or Gaussian distribution
% ***
SpHs = SpH*fss; % for the Wishart draw
SpHsc = chol(SpHs); % upper triangular where Sphsc' is lower triangular Choleski
%SpHsic = chol(inv(SpHs))'; % inverse Choleski decomposition -- lower triangular
A0hin = chol(SpH); % upper triangular, inverse of A0h, each column
% corresponds to an equation.
%----------------------------------------------
%
%@@@ The following can be used for other purposes such as forecasting
%
%swish = A0hin'; % each row corresponds to an equation
%A0h = inv(A0hin); % each column corresponds to an equation
%xa0 = A0h(a0indx);
%%*** form Bh (ncoef*nvar)
%Aplus = Hm1t*A0h; % estimate of Aplus -- the same as Astrar
%Bhp = Hm1t;
|