O mercado financeiro e inteligência artificial: uma comparação de modelos de aprendizagem automática com modelos tradicionais de precificação

Abstract

The aim of this work is to investigate the application of neural networks in op- tion pricing. Different architectures and parameters were tested in order to obtain better performance and greater prediction accuracy. Two volatility prediction mechanisms were used to compare the performance of the neural network with the parametric Black-Scholes model. Neural network performance was also compared with more advanced modular neural networks and a completely new technique was tested to add rational prediction hypotheses to neural networks, as well as the importance of adding virtual options that satisfy these hypotheses to better train the neural networks. In addition, the size and number of hidden layers were tested to provide a reference and a structured way of choo- sing neural network parameters. It was also proven that the prediction of neural networks surpasses traditional parametric methods.