Energy function construction and implementation for stock exchange prediction NNs

A. Cristea,T. Okamoto

DOI: https://doi.org/10.1109/KES.1998.726001

1998-04-21

Abstract:Neural networks (NN), with their parallel processing power, can be used as a tool to forecast stock exchange events (SEE), as a sub-domain of time-series (TS) forecasting. For the final product of SEE forecasts, other external economical factors have to be taken also into consideration and to be combined with the pure TS forecast. In this paper we present the energy function construction and implementation for SEE prediction. We focus on the mathematical deductions of the energy function and on the error minimization procedures. We present also some comparative results of our method, based on Lyapunov (also called infinite) norm, compared to the classical backpropagation method (BP), and to the random walk generator. We discuss some further optimisation of the system.