Name: RAFAEL CORREIA NASCIMENTO
Type: MSc dissertation
Publication date: 20/03/2018
Advisor:
Name | Role |
---|---|
ALBERTO FERREIRA DE SOUZA | Advisor * |
Examining board:
Name | Role |
---|---|
THIAGO OLIVEIRA DOS SANTOS | Internal Examiner * |
CLAUDINE SANTOS BADUE | Co advisor * |
ALBERTO FERREIRA DE SOUZA | Advisor * |
Summary: The use of automatic investment systems in the Brazilian Stock Exchange (BM&FBovespa) has been growing every year. This is because automatic investment systems, also called robots, are able to evaluate various financial assets simultaneously and at much shorter time scales than as of an investor. Thus, the need arises to create algorithms capable of analyzing large volumes of data in real time and making the decision on the best action to be taken for a particular financial asset of interest at any moment.
In this work, it was evaluated investment opportunities in the Brazilian futures market (a part of BM&FBovespa) in the time scale of tens of seconds, using an automatic investment system based on predictors and considering operation costs. Initially, it was evaluated the upper limit of return that can be generated by investments in the future market using a perfect predict, called oracle. Following, two types of neural predictors were evaluated: one based on Multilayer Perceptron neural networks (MLP) and the other based on VG-RAM weightless neural networks. Results showed that there are daily great investment opportunities in the time scales analyzed, but those were difficult to be predicted using the neural networks considered. This is because quotes of future market financial assets have a behavior very close to that of random-walk series. However, using decision mechanisms based on predictors recent performance, it is possible to improve the quality of buying and selling decisions, and to benefit from moments in which assets quotations series are more predictable.