Author(s):
Vieira, Cristiano Ribeiro
Date: 2013
Persistent ID: http://hdl.handle.net/10400.5/6340
Origin: Repositório da UTL
Subject(s): Artificial Neural Networks; Multilayer Perceptron; Backpropagation; Financial markets; Oil & Gas Industry; Forecasting.
Description
Mestrado em Matemática Financeira
Artificial Neural Networks are exible nonlinear mathematical models widely used in forecasting. This work is intended to investigate the support these models can give to nancial economists predicting prices movements of oil and gas companies listed in stock exchanges. Multilayer Perceptron models with logistic activation functions achieved better results predicting the direction of stocks returns than traditional linear regressions and better performances in companies with lower market capitalization. Furthermore, multilayer perceptron with eight hidden units in the hidden layer had better predictive ability than a neural network with four hidden neurons.