A dynamic modeling of stock prices and optimal decision making using MVP theory

Abstract

In this paper, first a precise mathematical model is obtained for four competing or cooperating companies' stock prices and then the optimal buy/sell signals are ascertained for five different agents which are trading in a virtual market and are trying to maximize their wealth over 1 trading year period. The model is so that gives a good prediction of the next 30th day stock prices. The companies used in this modeling are all chosen from Boston Stock Market. Genetic Programming (GP) is used to produce the predictive mathematical model. The interaction among companies and the effect imposed by each of five agents on future stock prices are also considered in our modeling. Namely, we have chosen eight companies in order that there is some kind of interrelation among them. Comparison of the GP models with Artificial Neural Networks (ANN) and Neuro-Fuzzy Networks (trained by the LoLiMoT algorithm) shows the superior potential of GP in prediction. Using these models; five players, each with a specific strategy and all with one common goal (wealth maximization), start to trade in a virtual market. We have also relaxed the short-sales constraint in our work. Each of the agents has a different objective function and all are going to maximize themselves. We have used Particle Swarm Optimization (PSO) as an evolutionary optimization method for wealth maximization. © 2009 Springer Netherlands.