An Epsilon Constraint Method for selecting Indicators for use in Neural Networks for Stock Market Forecasting

Abstract Forecasting future moves of stock markets has been and always will be of great interest to researchers and practitioners. This paper proposes a multi-objective programming methodology to select the optimum technical indicators to be used as input in a Neural Network (NN) in order to predict stock market prices. A new mathematical model will be proposed which involves objective functions and constraints to filter out the noisy signals and maximize the prediction power. The 0–1 multi-objective model aims to select the indicators maximizing the covariance of the indicators with the output of the NN while minimizing the covariance among the indicators themselves. The Multi-objective model is transformed via the Epsilon Constraint technique. Many efficient configurations of indicators for different values of epsilon are evaluated and their resulting errors are presented. Our approach provides a systematic methodology in order to choose the variables that significantly affect price movements. The methodology is applied on the NIKKEI225 stock market index.