Investigating neural network modeling decisions for the Australian All-Ordinaries Index

Abstract

Estimating stock market output depends mainly on identifying non-linear relationships of input variables. To forecast such systems a non-linear modeling tool is required. This paper describes the experimental approaches for developing an Artificial Neural Network for the purpose of modeling the Australian All Ordinaries Index movement over a prediction horizon of 1 year. Network parameters such as network architectures, input data sizing and periodicity are considered in the development of the network. The evaluation criterion for the Neural Network output is the R Square Statistic. © Springer-Verlag Berlin Heidelberg 2003.