177 lines
6.5 KiB
Matlab
177 lines
6.5 KiB
Matlab
function [LIK, likk,a,P] = univariate_kalman_filter(data_index,number_of_observations,no_more_missing_observations,Y,start,last,a,P,kalman_tol,riccati_tol,presample,T,Q,R,H,Z,mm,pp,rr,Zflag,diffuse_periods)
|
|
% Computes the likelihood of a stationnary state space model (univariate approach).
|
|
|
|
%@info:
|
|
%! @deftypefn {Function File} {[@var{LIK},@var{likk},@var{a},@var{P} ] =} univariate_kalman_filter (@var{data_index}, @var{number_of_observations},@var{no_more_missing_observations}, @var{Y}, @var{start}, @var{last}, @var{a}, @var{P}, @var{kalman_tol}, @var{riccati_tol},@var{presample},@var{T},@var{Q},@var{R},@var{H},@var{Z},@var{mm},@var{pp},@var{rr},@var{Zflag},@var{diffuse_periods})
|
|
%! @anchor{univariate_kalman_filter}
|
|
%! @sp 1
|
|
%! Computes the likelihood of a stationary state space model, given initial condition for the states (mean and variance).
|
|
%! @sp 2
|
|
%! @strong{Inputs}
|
|
%! @sp 1
|
|
%! @table @ @var
|
|
%! @item data_index
|
|
%! Matlab's cell, 1*T cell of column vectors of indices (in the vector of observed variables).
|
|
%! @item number_of_observations
|
|
%! Integer scalar, effective number of observations.
|
|
%! @item no_more_missing_observations
|
|
%! Integer scalar, date after which there is no more missing observation (it is then possible to switch to the steady state kalman filter).
|
|
%! @item Y
|
|
%! Matrix (@var{pp}*T) of doubles, data.
|
|
%! @item start
|
|
%! Integer scalar, first period.
|
|
%! @item last
|
|
%! Integer scalar, last period (@var{last}-@var{first} has to be inferior to T).
|
|
%! @item a
|
|
%! Vector (@var{mm}*1) of doubles, initial mean of the state vector.
|
|
%! @item P
|
|
%! Matrix (@var{mm}*@var{mm}) of doubles, initial covariance matrix of the state vector.
|
|
%! @item kalman_tol
|
|
%! Double scalar, tolerance parameter (rcond, inversibility of the covariance matrix of the prediction errors).
|
|
%! @item riccati_tol
|
|
%! Double scalar, tolerance parameter (iteration over the Riccati equation).
|
|
%! @item presample
|
|
%! Integer scalar, presampling if strictly positive (number of initial iterations to be discarded when evaluating the likelihood).
|
|
%! @item T
|
|
%! Matrix (@var{mm}*@var{mm}) of doubles, transition matrix of the state equation.
|
|
%! @item Q
|
|
%! Matrix (@var{rr}*@var{rr}) of doubles, covariance matrix of the structural innovations (noise in the state equation).
|
|
%! @item R
|
|
%! Matrix (@var{mm}*@var{rr}) of doubles, second matrix of the state equation relating the structural innovations to the state variables.
|
|
%! @item H
|
|
%! Vector (@var{pp}) of doubles, diagonal of covariance matrix of the measurement errors (corelation among measurement errors is handled by a model transformation).
|
|
%! @item Z
|
|
%! Matrix (@var{pp}*@var{mm}) of doubles or vector of integers, matrix relating the states to the observed variables or vector of indices (depending on the value of @var{Zflag}).
|
|
%! @item mm
|
|
%! Integer scalar, number of state variables.
|
|
%! @item pp
|
|
%! Integer scalar, number of observed variables.
|
|
%! @item rr
|
|
%! Integer scalar, number of structural innovations.
|
|
%! @item Zflag
|
|
%! Integer scalar, equal to 0 if Z is a vector of indices targeting the obseved variables in the state vector, equal to 1 if Z is a @var{pp}*@var{mm} matrix.
|
|
%! @item diffuse_periods
|
|
%! Integer scalar, number of diffuse filter periods in the initialization step.
|
|
%! @end table
|
|
%! @sp 2
|
|
%! @strong{Outputs}
|
|
%! @sp 1
|
|
%! @table @ @var
|
|
%! @item LIK
|
|
%! Double scalar, value of (minus) the likelihood.
|
|
%! @item likk
|
|
%! Column vector of doubles, values of the density of each observation.
|
|
%! @item a
|
|
%! Vector (@var{mm}*1) of doubles, mean of the state vector at the end of the (sub)sample.
|
|
%! @item P
|
|
%! Matrix (@var{mm}*@var{mm}) of doubles, covariance of the state vector at the end of the (sub)sample.
|
|
%! @end table
|
|
%! @sp 2
|
|
%! @strong{This function is called by:}
|
|
%! @sp 1
|
|
%! @ref{DsgeLikelihood}
|
|
%! @sp 2
|
|
%! @strong{This function calls:}
|
|
%! @sp 1
|
|
%! @ref{univariate_kalman_filter_ss}
|
|
%! @end deftypefn
|
|
%@eod:
|
|
|
|
% Copyright (C) 2004-2011 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 <http://www.gnu.org/licenses/>.
|
|
|
|
% AUTHOR(S) stephane DOT adjemian AT univ DASH lemans DOT fr
|
|
|
|
if nargin<20 || isempty(Zflag)% Set default value for Zflag ==> Z is a vector of indices.
|
|
Zflag = 0;
|
|
diffuse_periods = 0;
|
|
end
|
|
|
|
if nargin<21
|
|
diffuse_periods = 0;
|
|
end
|
|
|
|
% Get sample size.
|
|
smpl = last-start+1;
|
|
|
|
% Initialize some variables.
|
|
QQ = R*Q*transpose(R); % Variance of R times the vector of structural innovations.
|
|
t = start; % Initialization of the time index.
|
|
lik = zeros(smpl,1); % Initialization of the vector gathering the densities.
|
|
LIK = Inf; % Default value of the log likelihood.
|
|
oldP = Inf;
|
|
l2pi = log(2*pi);
|
|
notsteady = 1;
|
|
|
|
oldK = Inf;
|
|
K = NaN(mm,pp);
|
|
|
|
while notsteady && t<=last
|
|
s = t-start+1;
|
|
d_index = data_index{t};
|
|
if Zflag
|
|
z = Z(d_index,:);
|
|
else
|
|
z = Z(d_index);
|
|
end
|
|
oldP = P(:);
|
|
for i=1:rows(z)
|
|
if Zflag
|
|
prediction_error = Y(d_index(i),t) - z(i,:)*a;
|
|
Fi = z(i,:)*P*z(i,:)' + H(d_index(i));
|
|
else
|
|
prediction_error = Y(d_index(i),t) - a(z(i));
|
|
Fi = P(z(i),z(i)) + H(d_index(i));
|
|
end
|
|
if Fi>kalman_tol
|
|
if Zflag
|
|
Ki = (P*z(i,:)')/Fi;
|
|
else
|
|
Ki = P(:,z(i))/Fi;
|
|
end
|
|
if t>no_more_missing_observations
|
|
K(:,i) = Ki;
|
|
end
|
|
a = a + Ki*prediction_error;
|
|
P = P - (Fi*Ki)*Ki';
|
|
lik(s) = lik(s) + log(Fi) + prediction_error*prediction_error/Fi + l2pi;
|
|
end
|
|
end
|
|
a = T*a;
|
|
P = T*P*transpose(T) + QQ;
|
|
if t>=no_more_missing_observations
|
|
notsteady = max(abs(K(:)-oldK))>riccati_tol;
|
|
oldK = K(:);
|
|
end
|
|
t = t+1;
|
|
end
|
|
|
|
% Divide by two.
|
|
lik(1:s) = .5*lik(1:s);
|
|
|
|
% Call steady state univariate kalman filter if needed.
|
|
if t<last
|
|
[tmp, lik(s+1:end)] = univariate_kalman_filter_ss(Y,t,last,a,P,kalman_tol,T,H,Z,pp,Zflag);
|
|
end
|
|
|
|
% Compute minus the log-likelihood.
|
|
if presample
|
|
if presample>=diffuse_periods
|
|
lik = lik(1+(presample-diffuse_periods):end);
|
|
end
|
|
end
|
|
LIK = sum(lik); |