Jack Stephenson Undergraduate Dissertation 2015/16

A study into the application of Deep Neural Networks on Stock Market Trading

Supervised by E.Vasilaki

Abstract

The volatility of the stock market provides an interesting challengein maintaining a profit margin over time. Google DeepMind recently released a research letter\cite{Mnih2015} thatdescribes a method that shows human-level performance over a variety of Atari games. Using some of the insight and ideas from the research letter this project aims to createa learner that uses historical trading data, to maximise the profit when tradingon the stock market. The framework released by Google named TensorFlow has been chosen to undertake this proposal.This is due to a number of reasons, including the use by the Google department responsiblefor the research letter. This framework, and the algorithms underlying the research letter have shown great potentialwhen applied to a wide range of problems, including playing games as well as other tasks,such as image classification.This potential is the motivation for applying the techniques from the research paper to the fieldof stock trading.