Previsão do índice bursatil IBEX 35 usando redes neurais artificiais
Carregando...
Data
2021-04-09
Autores
Falcón Canillas, Salvador
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
A previsão de índices bolsistas de diferentes bolsas de valores é uma das questões mais importantes para economistas e investidores, a fim de conhecer, antecipadamente, os movimentos que ocorrem no mercado de investimento. Os investidores confiam, frequentemente, no seu conhecimento da economia dos mercados, dos fatores macroeconômicos dos países e de outros recursos, tais como: o comportamento dos gráficos de ações, para fazerem os seus investimentos da forma mais eficiente possível. O fato de serem pessoas que usam seu julgamento para investir no mercado de ações torna-se uma tarefa complicada, se não impossível, avaliar todos os fatores envolvidos no mercado de ações. Além dessa incapacidade, há fatores como elementos psicológicos e julgamentos errôneos subjetivos que podem levar o investidor a ter perdas econômicas, ou a não aproveitar boas oportunidades de investimento. Por essas razões, ao longo das últimas décadas, foram feitas tentativas de prever índices bolsistas, com diferentes sistemas entre eles: Modelos lineares, tais como ARIMA, Garch e Arch, redes neuronais tais como Feedforward Neural Network (FNN), Generalized Regression Neural Network (GRNN), Probabilistic Neural Network (PNN), etc. e a combinação de modelos lineares e redes neurais. Nesta pesquisa propõe-se uma forma como realizar a previsão do valor no dia seguinte do índice de ações espanhol IBEX 35, utilizando três redes neurais artificiais diferentes e comparando o desempenho de cada uma dessas redes em diferentes situações. O IBEX 35 é o principal índice de referência do mercado acionário da bolsa espanhola. É formado pelas 35 empresas mais líquidas listadas no Sistema de Interconexão da Bolsa de Valores Espanhola nas quatro bolsas de valores espanholas. As redes neurais a serem empregadas são Feedforward, Cascade-Forward e Generalized Regression
Forecasting stock market indices from different stock exchanges are one of the most important issues for economists and investors, to know in advance, the movements that occur in the investment market. Investors often rely on their knowledge of the economy of the markets, the macroeconomic factors of the countries, and other resources, such as the behaviour of stock charts, to make their investments as efficiently as possible. Being people who use their judgment to invest in the stock market is complicated, if not impossible, to evaluate all the factors involved in the stock market. In addition to this inability, there are factors such as psychological elements and subjective misjudgements that may lead the investor to have economic losses or not take advantage of good investment opportunities. For this reason, over the last decades, attempts have been made to predict stock market indices, with different systems among them: Linear models, such as ARIMA, Garch, and Arch, neural networks such as the Feedforward Neural Network (FNN), Generalized Regression Neural Network (GRNN), Probabilistic Neural Network (PNN), etc. and the combination of linear models and neural networks. This research presents how to forecast the next day value of the Spanish IBEX 35 stock index using three different artificial neural networks and comparing the performance of each of these nets in different situations. The IBEX 35 is the main reference index of the Spanish stock market. It is formed by the 35 most liquid companies listed on the Spanish Stock Exchange's Interconnection System on the four Spanish stock exchanges. The neural networks to use are Feedforward, Cascade-Forward, and Generalized Regression Neural Network.
Forecasting stock market indices from different stock exchanges are one of the most important issues for economists and investors, to know in advance, the movements that occur in the investment market. Investors often rely on their knowledge of the economy of the markets, the macroeconomic factors of the countries, and other resources, such as the behaviour of stock charts, to make their investments as efficiently as possible. Being people who use their judgment to invest in the stock market is complicated, if not impossible, to evaluate all the factors involved in the stock market. In addition to this inability, there are factors such as psychological elements and subjective misjudgements that may lead the investor to have economic losses or not take advantage of good investment opportunities. For this reason, over the last decades, attempts have been made to predict stock market indices, with different systems among them: Linear models, such as ARIMA, Garch, and Arch, neural networks such as the Feedforward Neural Network (FNN), Generalized Regression Neural Network (GRNN), Probabilistic Neural Network (PNN), etc. and the combination of linear models and neural networks. This research presents how to forecast the next day value of the Spanish IBEX 35 stock index using three different artificial neural networks and comparing the performance of each of these nets in different situations. The IBEX 35 is the main reference index of the Spanish stock market. It is formed by the 35 most liquid companies listed on the Spanish Stock Exchange's Interconnection System on the four Spanish stock exchanges. The neural networks to use are Feedforward, Cascade-Forward, and Generalized Regression Neural Network.
Descrição
Palavras-chave
Previsão, Ibex 35, Bolsa de valores, Backpropagation, Feedforward, Cascade, Regressão generalizada, Forecast, Stock market, Generalized regression