Análise de previsão de preços de ações de uma carteira otimizada, utilizando análise envoltória de dados, redes neurais artificiais e modelo de box-jenkins

Abstract

The stock market attracts more and more investors because of its high profitability, when compared to other investment options. However, this commercialization of assets has a high risk, and it is advisable to use mathematical concepts to assist in decision making. This paper aims to study some of these concepts applied in the stock exchange, to maximize the return and to control the risks. Firstly, for asset portfolio selection was used Data Envelopment Analysis (DEA), to then predict the stock price of the selected companies. For this prognosis, a comparison was made of time series forecast models, using the Artificial Neural Networks and Box-Jenkins method. For the DEA , it was used indicators such as price/profit, return, among others, as input and output variables. For the forecast, the daily commercial day price was used for each of the companies previously selected for the optimized portfolio. For the study with DEA, an optimized portfolio was obtained with a 15 times higher return than the Ibovespa index; and for the comparison of the forecasting methods, it was concluded that the Box-Jenkins methodology was the most adequate to predict share prices, due to lower error rates.