It's a little hard to second guess what you are trying to do, but it seems that you need to pad vector B with zeros or something (such as NaN) to make it the same length as the original A.
When you do the correlation now, you can just dump/ignore the final 300 values.
A = randn(1e3,1); % set some random data
S = randn(1e3,1); % set some more
L = length(A) - 300;
B = NaN(length(A), 1); % note B is pre-allocated same a A
for i = 1 : L
cc = corrcoef(A(i:i+300), S(i:i+300));
B(i) = cc(2,1);
end
plot(B);
Z=A-S;
figure(2)
plot(B,Z);
