[问答] Markov regime switching Model [推广有奖]

11楼

syasd 发表于 2012-6-16 21:31:07

楼主，我觉得“start gibbs toop”中的似然函数（第9——42行）可以改成这样，会和模型更符合一些
beta_by_s=zeros(n,1);
for t=1:n
if S(t)==1
   beta_by_s(t)=beta(1,1);
else
   if S(t)==2
      beta_by_s(t)=beta(2,1);
   else
      beta_by_s(t)=beta(3,1);
   end
end
end
err_by_s=y-beta_by_s;
h_by_s=zeros(n,1);
if S(1)==1
   h_by_s(1)=a_1(1)/(1-a_1(2)-a_1(3)-a_1(4));
else
   if S(2)==2
      h_by_s(1)=a_2(1)/(1-a_2(2)-a_2(3)-a_2(4));
   else
      h_by_s(1)=a_3(1)/(1-a_3(2)-a_3(3)-a_3(4));
   end
end

if S(2)==1
   h_by_s(2)=a_1(1)+a_1(2)*err_by_s(1)^2;
else
   if S(2)==2
      h_by_s(2)=a_2(1)+a_2(2)*err_by_s(1)^2;
   else
      h_by_s(3)=a_3(1)+a_3(2)*err_by_s(1)^2;
   end
end

if S(3)==1
   h_by_s(3)=a_1(1)+a_1(2)*err_by_s(2)^2+a_1(3)*err_by_s(1)^2;
else
   if S(3)==2
      h_by_s(3)=a_2(1)+a_2(2)*err_by_s(2)^2+a_2(3)*err_by_s(1)^2;
   else
      h_by_s(3)=a_3(1)+a_3(2)*err_by_s(2)^2+a_3(3)*err_by_s(1)^2;
   end
end

  for t=4:n
   if S(t)==1
      h_by_s(t)=a_1(1)+a_1(2)*err_by_s(t-1)^2+a_1(3)*err_by_s(t-2)+a_1(4)*err_by_s(t-3);
   else
      if S(t)==2
         h_by_s(t)=a_2(1)+a_2(2)*err_by_s(t-1)^2+a_2(3)*err_by_s(t-2)+a_2(4)*err_by_s(t-3);
      else
         h_by_s(t)=a_3(1)+a_3(2)*err_by_s(t-1)^2+a_3(3)*err_by_s(t-2)+a_3(4)*err_by_s(t-3);
      end
   end
  end

  lterm_by_s=(err_by_s.^2)./h_by_s;
  h_s1=h_by_s(S==1);
  h_s2=h_by_s(S==2);
  h_s3=h_by_s(S==3);
  lterm_s1=lterm_by_s(S==1);
  lterm_s2=lterm_by_s(S==2);
  lterm_s3=lterm_by_s(S==3);
  llike1=-0.5*sum(log(h_s1))-0.5*sum(lterm_s1);
  llike2=-0.5*sum(log(h_s2))-0.5*sum(lterm_s2);
  llike3=-0.5*sum(log(h_s3))-0.5*sum(lterm_s3);

我对MRS-GARCH了解有限，上面的程序是我自己的看法，不知道对不对。

我要一步一步往上爬，在最高点乘着叶片往前飞，任风吹干流过的泪和汗，总有一天我有属于我的天。