Application of artificial neural networks to predict the behavior of stocks

José Ricardo Magalhães Rivero, Cleber Almeida Corrêa Junior, Rosilene Abreu Portella Corrêa

Stocks, Artificial Neural Networks, Multi-Layer Perceptron with Backpropagation, Probability of series behavior.

Statistical data point to the fact that the vast majority of the world population, even after working for a lifetime, when they retire, do not have significant reserves of financial resources in order to guarantee a good quality of life in the elderly. Bearing in mind that the financial stock market offers a viable opportunity for lifelong capital expansion; Through this work, we sought to develop an innovative technique to allow a simple support based on mathematical models, for support decision making by common people, for buying or selling market stocks. This is because the techniques that support decision-making are relatively complex and not widely mastered by the majority of the Brazilian population. The algorithm was proposed to better perform this task. It was made by one corresponding Artificial Neural Network of the “Multi-Layer Perceptron” type with “Backpropagation”. Because this ANN is suitable for learning patterns of historical series, which are usually the object of study of stock price behavior by technical analysis methodologies, that are widely used by the market. Therefore, a comparative study was carried out between the results found using the proposed ANN methodology versus the results obtained from simple technical analysis versus single purchase and sale operations in a period of one year. It was found that the ANN model used guided the achievement of superior results for operations with all the Stocks tested, thus proving to be a promising way to solve problems of this nature; related to the identification of mathematical patterns of historical series of the behavior of stock prices on the São Paulo stock exchange.

Received: 26 Jan 2023, Receive in revised form:25 Feb 2023, Accepted: 03 Mar 2023, Available online: 13 Mar 2023

10.22161/ijaers.103.1