Help with Kalman filter with S&P 500 index

I am using S&P 500 index time series and using Kalman filter to fit it using the combination of weights * 3 previous index values (y-3, y-2, y-1). I am doing it recursively and updating the estimation of the weights at each step. The weights should converge to a final value, faster or slower depending on the ratio of W and V error signal. When I test different values of V and W it makes no difference to the speed of convergence of the parameters. The code is below:

I = eye(W);

flag = 0;

     for i = W+1:(size(data))
         %generate X
         for window = 1:W
            X(i-W,window) = data(i-(W+1-window));
         end
            %set theta and P or use previous value
            if (flag == 0)
                theta = t ;
                P = alpha*I;
                flag = 1;
            else
                P = P;
            end

            %calculate prediction
            x_tmp = X(i-W,:)';
            y_hat = theta' * x_tmp;
            %gather the weights and the Y's
            weights(i-W,:) = theta;
            estimated(i-W,:) = y_hat;

            %calc error
            e = data(i) - y_hat;
            err(i-W,:) = e;

            %calc Kalman gain
            K = (P*x_tmp)/(2500 + x_tmp'*P*x_tmp);
            %update theta and P
            theta = theta + K*e;
            P = (I - K*x_tmp')*P;

     end
  end

[Walter Roberson]

duplicates http://uk.mathworks.com/matlabcentral/answers/216296-help-me-to-find-a-bug

[Jacek Freyer]

I know but I still have no answer:)