ffnet9fun_jump.m

Neural Network in Finance (神经网络在金融界:赢得预言性的优势)全部原码。内容包括预测与估计

function [criterion,sse3,g,A3,W3,b3,W4,b4,W5,b5,W6,b6, W7, b7] = ffnet9fun_jump(beta2);
global P T nlayer nneuron1 nneuron2 nneuron3;
[rt ct] = size(T); rt = min([rt ct]);
[rp cp] = size(P);
cp1 = (cp+1) * ct;
beta = beta2(1:end-cp1);
W5 = 0; b5 = 0; W6 = 0; b6 = 0;
if nlayer == 1, 
beta1 = beta(1:nneuron1*cp);
W3 = reshape(beta1,cp,nneuron1);
b3 = beta(nneuron1*cp+1:nneuron1*cp+nneuron1);
W4 = beta(nneuron1*cp+nneuron1+1:end-ct);
W4 = reshape(W4,nneuron1,ct);
b4 = beta(end-ct+1:end);
W5 = beta2(end-cp1+1:end-ct);
W5 = reshape(W5,cp,ct);
b5 = beta2(end-ct+1:end);
W6 = [];
b6 = [];
W7 = [];
b7 = [];
% A3 = simuff(P,W3,b3,'logsig',W4,b4,'purelin');
A31 =   P  * W3 +  kron(ones(length(P),1),b3);
A32  =   1 ./ (1 + exp(-A31));
A33  =  A32  * W4  +   kron(ones(length(A32),1),b4);
A3 = A33 + P * W5 + kron(ones(length(P),1),b5);
elseif nlayer == 2,
beta1 = beta(1:nneuron1*cp);
W3 = reshape(beta1,cp, nneuron1);
b3 = beta(nneuron1*cp+1:nneuron1*cp+nneuron1);
beta2 = beta(nneuron1*cp+nneuron1+1:nneuron1*cp+nneuron1+nneuron1*nneuron2);
W4 = reshape(beta2,nneuron1, nneuron2);
b4 = beta(nneuron1*cp+nneuron1+nneuron1*nneuron2+1:nneuron1*cp+nneuron1+nneuron1*nneuron2+nneuron2);
W5 = beta(nneuron1*cp+nneuron1+nneuron1*nneuron2+nneuron2+1:end-ct);
W5 = reshape(W5,nneuron2,rt);
b5 = beta(end-ct+1:end);
W6 = beta2(end-cp1+1:end-ct);
W6 = reshape(W5,cp,ct);
b6 = beta2(end-ct+1:end);
W7 = [];
b6 = [];
% A3 = simuff(P,W3, b3, 'logsig', W4, b4, 'logsig', W5, b5, 'purelin');
A31 =   P  * W3 +  kron(ones(length(P),1),b3);
A32  =   1 ./ (1 + exp(-A31));
A33 =  A32 * W4 + kron(ones(length(A32),1),b4);
A34  =   1 ./ (1 + exp(-A33));
A35 =  A34 * W5 + kron(ones(length(A34),1),b5);
A3 = A35;
A3 = A3 + P * W6 +  kron(ones(length(P),1),b6);
else
beta1 = beta(1:nneuron1*cp);
W3 = reshape(beta1,cp, nneuron1);
b3 = beta(nneuron1*cp+1:nneuron1*cp+nneuron1);
beta2 = beta(nneuron1*cp+nneuron1+1:nneuron1*cp+nneuron1+nneuron1*nneuron2);
W4 = reshape(beta2, nneuron1, nneuron2);
b4 = beta(nneuron1*cp+nneuron1+nneuron1*nneuron2+1:nneuron1*cp+nneuron1+nneuron1*nneuron2+nneuron2);
beta3 = beta(nneuron1*cp+nneuron1+nneuron1*nneuron2+nneuron2+1:nneuron1*cp+nneuron1+nneuron1*nneuron2+nneuron2+ nneuron2* nneuron3);
W5 = reshape(beta3, nneuron2, nneuron3);
b5 = beta(nneuron1*cp+nneuron1+nneuron1*nneuron2+nneuron2+ nneuron2* nneuron3+1:nneuron1*cp+nneuron1+nneuron1*nneuron2+nneuron2+ nneuron2* nneuron3+nneuron3);
W6 = beta(nneuron1*cp+nneuron1+nneuron1*nneuron2+nneuron2+ nneuron2* nneuron3+nneuron3+1:end-ct);
W6 = reshape(W6,nneuron3,ct);
b6 = beta(end-ct+1:end);
W7 = beta2(end-cp1+1:end-ct);
W7 = reshape(W5,cp,ct);
b7 = beta2(end-ct+1:end);
A31 =  P * W3 + kron(ones(length(P),1),b3);
A32  =   1 ./ (1 + exp(-A31));
A33 =  A32 * W4 + kron(ones(length(A32),1),b4);
A3  =   1 ./ (1 + exp(-A3));
NPC = A3;
A3 =  A3 * W5 + kron(ones(length(A3),1),b5);
A3  =   1 ./ (1 + exp(-A3));
A3 =   A3*W6 +  kron(ones(length(A3),1),b6);
A3 = A3 + P * W7 + kron(ones(length(P),1),b3);
end
T = T;
A3 = A3; A3 = real(A3);
TT = T;
AA3 = A3;
err3 = TT- AA3;
sse3 =  sum(err3 .^2);
sigmaerr = inv(cov(err3));
criterion = sse3 * sigmaerr * sse3';
criterion = real(criterion);
criterion = real(criterion);
ssrsq3 = var(A3) ./ var(T);
ssrsq3 = ssrsq3;
g = [];