This preview shows page 1. Sign up to view the full content.
Unformatted text preview: r the
short term module. The authors conclude that the neural network shows a much better
response than multiple linear regression.
Neural sequential associator
In this paper , the author uses a feedforward neural network with the last n stock
index values as inputs and the next N-n values as the outputs. This is a N-n step ahead
prediction. Thus, if index for the nth day is denoted by Xn, then, the inputs are X1, X2, ... Xn
and the outputs are Xn+1, Xn+2, ..., XN. If such a network is trained, any correlation
between the index values for the n+1 through Nth day will be neglected. To ensure that
this does not happen, the network is trained with errors between the desired and actual
outputs in addition to the n inputs. These errors will then be (Xn+1 - Yn+1) ...., where Y is
the output of the network. As training proceeds this error will tend to zero and these
additional inputs are not required in the testing phase. This work also uses two neural
networks, one to learn the global features and another to learn the local feature...
View Full Document
- Spring '12