Daily Ecopertrol stock performance estimate by using estimation of distribution algorithms (EDAs): Evolutionary computation

In this paper a model, which is based on the principles of evolutionary computation, will be developed and analyzed for financial forecasting and time series prediction. The model refers to returns, prices and transaction volumes of stocks that are listed on the Colombian Stock Exchange (BVC), and particularly to time series that represent the returns of the stocks of Ecopetrol. The proposed model uses the estimation of distribution algorithms (EDA's) for the short-term projection of the returns of such stocks. The EDAs, unlike other optimization techniques that are based on traditional evolution algorithms, provide greater statistical robustness in shaping the population of individuals and with respect to the number of generations that are required to solve the problem. For this purpose the EDA incorporates three dynamic populations that explore the solution space of the problem by using a mechanism which propagates the differences that exist among the best individuals. Additionally, it includes a strategy to preserve the diversity of the population. The proposed model allowed short-term projections for the future values of the returns of Ecopetrol stocks, departing from a set of parameters and variables that make up the stock price. This way the behavior of a specific asset can be described over time and the complexity which is associated with the application of traditional techniques, that allow the modeling of financial time series, can be eliminated.