function [endogenousvariables, exogenousvariables] = backward_model_inversion(constraints, exogenousvariables, initialconditions, endo_names, exo_names, freeinnovations, DynareModel, DynareOptions, DynareOutput)

% INPUTS
% - constraints         [dseries]        with N constrained endogenous variables from t1 to t2.
% - exogenousvariables  [dseries]        with Q exogenous variables.
% - initialconditions   [dseries]        with M endogenous variables starting before t1 (M initialcond must contain at least the state variables).
% - endo_names          [cell]           list of endogenous variable names.
% - exo_names           [cell]           list of exogenous variable names.
% - freeinstruments     [cell]           list of exogenous variable names used to control the constrained endogenous variables.
%
% OUTPUTS
% - endogenous          [dseries]
% - exogenous           [dseries]
%
% REMARKS

% Copyright (C) 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 <http://www.gnu.org/licenses/>.

% Get indices for the calibrated and free innovations.
freeinnovations_id = zeros(length(freeinnovations), 1);
if length(freeinnovations)<DynareModel.exo_nbr
    for i=1:length(freeinnovations)
        freeinnovations_id(i) = strmatch(freeinnovations{i}, exo_names, 'exact');
    end
    calibratedinnovations_id = setdiff(transpose(1:length(exo_names)), freeinnovations_id);
else
    freeinnovations_id = transpose(1:length(exo_names));
    calibratedinnovations_id = [];
end

nxfree = length(freeinnovations_id);
nxcalb = length(calibratedinnovations_id);

% Get indices for the the controlled and free endogenous variables.
controlledendogenousvariables_id = zeros(length(freeinnovations), 1);
if length(freeinnovations)<DynareModel.endo_nbr
    for i=1:length(freeinnovations)
        controlledendogenousvariables_id(i) = strmatch(constraints.name{i}, endo_names, 'exact');
    end
    freeendogenousvariables_id = setdiff(transpose(1:length(endo_names)), controlledendogenousvariables_id);
else
    controlledendogenousvariables_id = transpose(1:length(endo_names));
    freeendogenousvariables_id = [];
end

nyfree = length(freeendogenousvariables_id);
nyctrl = length(controlledendogenousvariables_id);

% Get indices of variables appearing at time t-1.
iy1 = find(DynareModel.lead_lag_incidence(1,:)>0);

% Get indices of variables appearing at time t.
iy0 = find(DynareModel.lead_lag_incidence(2,:)>0);

% Set indices for trust_region algorithm.
idx = 1:DynareModel.endo_nbr;
jdx = 1:(nyfree+nxfree);

% Build structure to be passed to the objective function.
ModelInversion.nyfree = nyfree;
ModelInversion.nyctrl = nyctrl;
ModelInversion.nxfree = nxfree;
ModelInversion.nxcalb = nxcalb;
ModelInversion.y_constrained_id = vec(DynareModel.lead_lag_incidence(2,controlledendogenousvariables_id));
ModelInversion.y_free_id = vec(DynareModel.lead_lag_incidence(2,freeendogenousvariables_id));
ModelInversion.x_free_id = freeinnovations_id;
ModelInversion.J_id = [ModelInversion.y_free_id ; sum(DynareModel.lead_lag_incidence(:)>0)+ModelInversion.x_free_id];

% Get the name of the dynamic model routines.
model_dynamic = str2func([DynareModel.fname,'_dynamic']);
model_dtransf = str2func('dynamic_backward_model_for_inversion');

% Initialization of vector y (free endogenous variables and free innovations).
y = NaN(nyfree+nxfree);

% Initialization of the returned simulations (endogenous variables).
Y = NaN(DynareModel.endo_nbr, nobs(constraints)+1);
initialconditions
constraints.dates(1)
Y(:,1) = initialconditions(constraints.dates(1)-1).data(1:DynareModel.endo_nbr);
for i=1:nyctrl
    Y(controlledendogenousvariables_id(i),2:end) = transpose(constraints.data(:,i));
end

% Initialization of the returned simulations (exogenous variables).
X = exogenousvariables{exo_names{:}}(constraints.dates(1)-max(1,DynareModel.maximum_exo_lag):constraints.dates(end)).data;

% Inversion of the model, solvers for the free endogenous and exogenous variables (call a Newton-like algorithm in each period).
for it = 2:nobs(constraints)+1
    % Set the lagged values of the endogenous variables.
    ylag = Y(iy1,it-1);
    % Set the current values of the constrained endogenous variables.
    ycur = Y(controlledendogenousvariables_id,it);
    % Vector z gather the free endogenous variables (initialized with lagged
    % values) and the free exogenous variables (initialized with 0).
    z = [Y(freeendogenousvariables_id,it-1); zeros(nxfree, 1)];
    % Solves for z.
    z = dynare_solve(model_dtransf, z, DynareOptions, model_dynamic, ylag, ycur, X, DynareModel.params, DynareOutput.steady_state, it+DynareModel.maximum_exo_lag, ModelInversion);
    % Update the matrix of exogenous variables.
    X(it,freeinnovations_id) = z(nyfree+1:end);
    % Update the matrix of endogenous variables.
    Y(freeendogenousvariables_id,it) = z(1:nyfree);
end

endogenousvariables = dseries(Y', constraints.dates(1)-1, endo_names);
exogenousvariables = dseries(X(max(DynareModel.maximum_exo_lag,1)+1:end,:), constraints.dates(1), exo_names);