Global Decision Strategy based on combining multiple regression methods

Abstract

One of the most used prediction and decision-making methods is the approximation through regression methods. However, ordinary regression methods often perform poorly in real world applications. For this reason, many researches have been done in least squares combinations « Granger and Ramanathan 1984) », and regression-based forecast combination « Coulson and Robins (1993), Deutsch et al.(1994) » are an important development in the forecast combination literature. In this respect we propose a new approach for regression based on combining simultaneously multiple regression methods in order to achieve better predictions than ordinary regression techniques. It is a clear fact that a particular approach can’t be highly appreciated unless its efficiency is practically valid, hence, we have applied our strategy to predict CAC-40 evolution. In fact, our results show a better efficiency than all used ordinary regression methods.