File size: 5,644 Bytes
ed7d493 | 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 | clear,clc
%---------------------------------------------------------
% Add folder 'files' to the search path and load the model
%---------------------------------------------------------
addpath('files');
load('model')
%----------------------------------------------------------------------------
% Provide nodes and weights for the quadrature that approximates expectations
%----------------------------------------------------------------------------
n_e=1; % number of shocks.
[n_nodes,nodes,weights] = Monomials_2(n_e,eye(n_e)); % this quadrature function was written by Judd, Maliar, Maliar and Valero (2014).
nodes=nodes'; % transpose to n_e-by-n_nodes
%----------------------------------
% Make a vector of parameter values
%----------------------------------
BETA=.96; GAMMA=2; ALPHA=.3; RHO=.8; DELTA=.1; SIGMA=.02;
params=eval(symparams);
%----------------------------------------------------------------------
% Prepare an initial guess - in this case I use a perturbation solution
%----------------------------------------------------------------------
% Steady state values
kss=((1/BETA-1+DELTA)/ALPHA)^(1/(ALPHA-1));
zss=0;
css=kss^ALPHA-DELTA*kss;
nxss=[kss;zss];
nyss=css;
% Cross moments of the shocks
M=get_moments(nodes,weights,model.order(2));
% Compute the perturbation solution (keep the 4 outputs):
[derivs,stoch_pert,nonstoch_pert,model]=get_pert(model,params,M,nxss,nyss);
% Explanation of outputs:
% derivs=structure with the perturbation solution as explained in Levintal
% (2017): "Fifth-Order Perturbation Solution to DSGE Models".
% stoch_pert=the perturbation solution in the form of unique polynomial coefficients.
% nonstoch_pert=same as stoch_pert but without correction for the model volatility (i.e. this is a perturbation solution of a deterministic version of the model)
%-------------------------------------
% Solve the model by Taylor projection
%-------------------------------------
x0=nxss; % the approximation point (here we use the steady state, but it could be any arbitrary state)
c0=nxss; % the center of the initial guess
% tolerance parameters for the Newton solver
tolX=1e-6;
tolF=1e-6;
maxiter=10;
% model.jacobian='exact'; % this is the default
% model.jacobian='approximate'; % for large models try the approximate jacobian.
initial_guess=stoch_pert;
[coeffs,model]=tpsolve(initial_guess,x0,model,params,c0,nodes,weights,tolX,tolF,maxiter);
%------------------------------------------------------------------
% Compute the residual function and the model variables at point x0
%------------------------------------------------------------------
[R_fun0,g_fun0,Phi_fun0,auxvars0]=residual(coeffs,x0,params,c0,nodes,weights);
% R_fun0 is the residual function at x0.
% g_fun0 is the control variables at x0, namely, g(x0).
% Phi_fun0 is the function Phi at x0 and each future node, namely, Phi(x0,g(x0),epsp), for each node of the quadrature.
% auxvars0 is the auxiliary functions at x0 and each future node.
% compute the function g(x) at x0
y0=evalg(x0,coeffs,c0);
% compute the function Phi(x,y,epsp) at x0, y0 and epsp0
epsp0=.02;
xp0=evalPhi(x0,y0,epsp0,params);
%---------------------------------
% simulate the model for T periods
%---------------------------------
T=100;
shocks=randn(1,T+1); % draw shocks
% preallocate
x_simul=zeros(model.n_x,T+1);
y_simul=zeros(model.n_y,T);
R_simul=zeros(model.n_y,T);
x_simul(:,1)=x0;
% option=1; % compute only simulated variables
option=2; % compute model residuals
for t=1:T
xt=x_simul(:,t);
epsp=shocks(t+1);
% Option 1 - compute only the simulated variables
if option==1
yt=evalg(xt,coeffs,c0);
y_simul(:,t)=yt;
x_simul(:,t+1)=evalPhi(xt,yt,epsp,params);
else
% Option 2 - compute also model residuals
[Rt,yt]=residual(coeffs,xt,params,c0,nodes,weights);
y_simul(:,t)=yt;
x_simul(:,t+1)=evalPhi(xt,yt,epsp,params);
R_simul(:,t)=Rt;
end
end
%-------------------------------------------
% Solve the model again at a different state
%-------------------------------------------
% This is useful when the long run domain of the model is far from the
% initial state, so we need to approximate the solution at the long run state
% (e.g. the risky steady state or the mean of the ergodic distribution)
% rather than the steady state.
x1=x0*1.1; % take some arbitrary state
[coeffs1,model]=tpsolve(coeffs,x1,model,params,c0,nodes,weights,tolX,tolF,maxiter); % solve at x1
%-----------------------
% Use a different solver
%-----------------------
% The function tpsolve uses the Newton method for up to maxiter iterations. If it fails, it
% switches automatically to fsolve for another maxiter iterations. You can
% control the parameters of the second solver by optimoptions. The
% supported solvers are fsolve and lsqnonlin.
% For example, do one Newton iteration and switch to lsqnonlin:
x2=x1*1.1;
maxiter=1; % one Newton iteration
OPTIONS = optimoptions('lsqnonlin','TolX',tolX,'TolF',tolF,'MaxIter',10,'Display','iter-detailed'); % 10 more iterations by lsqnonlin
[coeffs2,model]=tpsolve(coeffs,x2,model,params,c0,nodes,weights,tolX,tolF,maxiter,OPTIONS);
% or switch to fsolve:
maxiter=1; % one Newton iteration
OPTIONS = optimoptions('fsolve','TolX',tolX,'TolF',tolF,'MaxIter',10,'Display','iter-detailed'); % 10 more iterations by fsolve
[coeffs3,model]=tpsolve(coeffs,x2,model,params,c0,nodes,weights,tolX,tolF,maxiter,OPTIONS);
|