markov process and dynamic programming

Lima Vilchez

2013 6 13

0 个回答

2 次查看（30 天）

0 个投票

Im relatively new in Matlab, and im having some problems when using finite horizon dynamic programming while using 2 state variables,one of which follows a Markov process. This is as far as ive gone:

r=0.04;
B=100/106;
T=60;
delta=2;
na=150;
ny=2;
b=-1; % 
amin=b;
amax=5;
rho=0.95;
sigma=0.5; %desv std del proceso
sizes=1;     
Pi=Tauchen(rho,sigma,ny,sizes); % I worked this on some other file
pi=Pi(1:ny,:); 
yg=exp(Pi(ny+1,:)); 
ag=linspace(amin,amax,na); 
lag=linspace(log(amin+1),log(amax+1),na)';
ag=exp(lag)-1; 
%Up to this part, its ok
V=zeros(na,ny,T); %
VT1=zeros(na,ny,T); 
Vms=zeros(na,1,T); 
G=zeros(na,ny,T);
for t=1:T
    for m=1:na
        for s=1:ny
            for j=1:na
                if yg(s)+(1+r)*ag(m)-ag(j)<0; 
                Vms(j,1,t)=-1e10; 
                else
                Vms(j,1,t)=((yg(s)+(1+r)*ag(m)-ag(j))^(1-delta)-1)/(1-delta)+ B*pi(s,:)*VT1(j,:,t)'; 
                end
            end
            [h_vf h_pf]=max(Vms); 
            V(m,s,t)=h_vf;
            G(m,s,t)=ag(h_pf);
        end
    end
VT1=V;
end

it seems matrix sizes dont match, but im not sure if this is the way to solve a finite horizon problem. with a infinite horizon case, i used a "while"(instead of looping the time variable) until it converged and it worked just fine. Id really appreciate it if someone shed some light on this!