Meta-heurísticas aplicadas ao problema de projeção do preço de ações na bolsa de valores

Abstract

The stock prices prediction in the stock exchange is an attractive field for research due to its commercial applications and financial benefits offered. The objective of this work is to analyze the performance of two meta-heuristic algorithms, Bat Algorithm and Genetic Algorithm to the problem of stock prices prediction. The individuals in the population of the algorithms were modeled using 7 technical indicators. The profit at the end of a period is maximized by choosing the right time to buy and sell stocks. To evaluate the proposed methodology, experiments were performed using real historical data (2006-2012) of 92 stocks listed on the stock exchange in Brazil. Cross-validation was applied in the experiments to avoid the overfiting using 3 periods for training and 4 for testing. The results of the algorithms were compared among them and also the performance indicator BuyandHold (B&H).For 91.30% of the stocks, the algorithms obtained profit higher than the B&H, and in 79.35% of them Bat Algorithm had the best performance, while for 11.95% of the stocks Genetic Algorithm was better. The results indicate that it is promising to apply meta-heuristics with the proposed model to the problem of stock prices prediction in the stock exchange.