sequential_importance_particle_filter.m 6.52 KB
Newer Older
1
2
function [LIK,lik] = sequential_importance_particle_filter(ReducedForm,Y,start,DynareOptions)
% Evaluates the likelihood of a nonlinear model with a particle filter (optionally with resampling).
Frédéric Karamé's avatar
Frédéric Karamé committed
3
4
5
% Standard Sequential Monte Carlo approach with 
%    - the usual proposal (the state transition distribution)
%    - options on resampling: none, adaptive or systematic 
6
7
8
9
%@info:
%! @deftypefn {Function File} {@var{y}, @var{y_} =} sequential_importance_particle_filter (@var{ReducedForm},@var{Y}, @var{start}, @var{DynareOptions})
%! @anchor{particle/sequential_importance_particle_filter}
%! @sp 1
10
%! Evaluates the likelihood of a nonlinear model with a particle filter (optionally with resampling).
11
12
13
14
15
16
17
18
%!
%! @sp 2
%! @strong{Inputs}
%! @sp 1
%! @table @ @var
%! @item ReducedForm
%! Structure describing the state space model (built in @ref{non_linear_dsge_likelihood}).
%! @item Y
19
%! p*smpl matrix of doubles (p is the number of observed variables), the (detrended) data.
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
%! @item start
%! Integer scalar, likelihood evaluation starts at observation 'start'.
%! @item DynareOptions
%! Structure specifying Dynare's options.
%! @end table
%! @sp 2
%! @strong{Outputs}
%! @sp 1
%! @table @ @var
%! @item LIK
%! double scalar, value of (minus) the logged likelihood.
%! @item lik
%! smpl*1 vector of doubles, density of the observations at each period.
%! @end table
%! @sp 2
%! @strong{This function is called by:}
%! @ref{non_linear_dsge_likelihood}
%! @sp 2
%! @strong{This function calls:}
%!
%! @end deftypefn
%@eod:

Houtan Bastani's avatar
Houtan Bastani committed
43
% Copyright (C) 2011-2013 Dynare Team
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
%
% 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/>.

60
% AUTHOR(S) frederic DOT karame AT univ DASH lemans DOT fr
61
%           stephane DOT adjemian AT univ DASH lemans DOT fr
62

63
64
persistent init_flag
persistent mf0 mf1
65
persistent number_of_particles number_of_state_variables
66
67
68
69
70
71
72
persistent sample_size number_of_observed_variables number_of_structural_innovations

% Set default value for start
if isempty(start)
    start = 1;
end

73
74
75
% Set flag for prunning
pruning = DynareOptions.particle.pruning;

76
% Get steady state and mean.
Frédéric Karamé's avatar
Frédéric Karamé committed
77
steadystate = ReducedForm.steadystate;
78
79
80
constant = ReducedForm.constant;
state_variables_steady_state = ReducedForm.state_variables_steady_state;

81
% Set persistent variables (if needed).
82
83
84
85
if isempty(init_flag)
    mf0 = ReducedForm.mf0;
    mf1 = ReducedForm.mf1;
    sample_size = size(Y,2);
86
    number_of_state_variables = length(mf0);
87
    number_of_observed_variables = length(mf1);
88
    number_of_structural_innovations = length(ReducedForm.Q);
89
90
91
92
93
94
95
96
97
98
99
100
101
    number_of_particles = DynareOptions.particle.number_of_particles;
    init_flag = 1;
end

% Set local state space model (first order approximation).
ghx  = ReducedForm.ghx;
ghu  = ReducedForm.ghu;

% Set local state space model (second order approximation).
ghxx = ReducedForm.ghxx;
ghuu = ReducedForm.ghuu;
ghxu = ReducedForm.ghxu;

102
% Get covariance matrices.
103
Q = ReducedForm.Q; % Covariance matrix of the structural innovations.
Stéphane Adjemian's avatar
Stéphane Adjemian committed
104
H = ReducedForm.H; % Covariance matrix of the measurement errors.
105
106
107
108
if isempty(H)
    H = 0;
end

109
110
111
112
% Initialization of the likelihood.
const_lik = log(2*pi)*number_of_observed_variables;
lik  = NaN(sample_size,1);

113
114
115
% Get initial condition for the state vector.
StateVectorMean = ReducedForm.StateVectorMean;
StateVectorVarianceSquareRoot = reduced_rank_cholesky(ReducedForm.StateVectorVariance)';
116
117
118
119
120
121
if pruning
    StateVectorMean_ = StateVectorMean;
    StateVectorVarianceSquareRoot_ = StateVectorVarianceSquareRoot;
end

% Get the rank of StateVectorVarianceSquareRoot
122
state_variance_rank = size(StateVectorVarianceSquareRoot,2);
123
124

% Factorize the covariance matrix of the structural innovations
125
126
127
Q_lower_triangular_cholesky = chol(Q)';

% Set seed for randn().
128
set_dynare_seed('default');
129
130

% Initialization of the weights across particles.
131
weights = ones(1,number_of_particles)/number_of_particles ;
132
StateVectors = bsxfun(@plus,StateVectorVarianceSquareRoot*randn(state_variance_rank,number_of_particles),StateVectorMean);
133
134
135
136
137
if pruning
    StateVectors_ = StateVectors;
end

% Loop over observations
138
139
140
for t=1:sample_size
    yhat = bsxfun(@minus,StateVectors,state_variables_steady_state);
    epsilon = Q_lower_triangular_cholesky*randn(number_of_structural_innovations,number_of_particles);
141
142
143
144
145
146
    if pruning
        yhat_ = bsxfun(@minus,StateVectors_,state_variables_steady_state);
        [tmp, tmp_] = local_state_space_iteration_2(yhat,epsilon,ghx,ghu,constant,ghxx,ghuu,ghxu,yhat_,steadystate,DynareOptions.threads.local_state_space_iteration_2);
    else
        tmp = local_state_space_iteration_2(yhat,epsilon,ghx,ghu,constant,ghxx,ghuu,ghxu,DynareOptions.threads.local_state_space_iteration_2);
    end
147
    PredictedObservedMean = tmp(mf1,:)*transpose(weights);
148
149
    PredictionError = bsxfun(@minus,Y(:,t),tmp(mf1,:));
    dPredictedObservedMean = bsxfun(@minus,tmp(mf1,:),PredictedObservedMean);
150
    PredictedObservedVariance = bsxfun(@times,dPredictedObservedMean,weights)*dPredictedObservedMean' + H;
151
152
153
    lnw = -.5*(const_lik+log(det(PredictedObservedVariance))+sum(PredictionError.*(PredictedObservedVariance\PredictionError),1));
    dfac = max(lnw);
    wtilde = weights.*exp(lnw-dfac);
154
    lik(t) = log(sum(wtilde))+dfac;
155
    weights = wtilde/sum(wtilde);
156
157
    if (strcmp(DynareOptions.particle.resampling.status,'generic') && neff(weights)<DynareOptions.particle.resampling.neff_threshold*sample_size ) || ...
        strcmp(DynareOptions.particle.resampling.status,'systematic')
158
        if pruning
159
160
161
162
            temp = resample([tmp(mf0,:)' tmp_(mf0,:)'],weights,DynareOptions);
            StateVectors = temp(:,1:number_of_state_variables)' ;
            StateVectors_ = temp(:,number_of_state_variables+1:2*number_of_state_variables)';
        else
Frédéric Karamé's avatar
Frédéric Karamé committed
163
            StateVectors = resample(tmp(mf0,:)',weights',DynareOptions)';
164
        end
Stéphane Adjemian's avatar
Stéphane Adjemian committed
165
        weights = ones(1,number_of_particles)/number_of_particles;
Stéphane Adjemian's avatar
Stéphane Adjemian committed
166
    elseif strcmp(DynareOptions.particle.resampling.status,'none')
167
        StateVectors = tmp(mf0,:);
168
        if pruning
Frédéric Karamé's avatar
Frédéric Karamé committed
169
            StateVectors_ = tmp_(mf0,:);
170
        end
171
172
173
    end
end

174
LIK = -sum(lik(start:end));