An Introduction to Deep Reinforcement Learning

This degree work aims to explore the use of synthetic financial time series generated by a Generative Adversarial Neural Networks (GAN) model to train a Deep Reinforcement Learning algorithm that executes buy and sell actions for a stock in the Standard & Poor's 500 index.For the implementation of the study, we used the CRISP methodology proposed by IBM, understanding first the business and the theory necessary to develop the models, to continue with the exploration and knowledge of the available data that matched the objectives of the project.In this paper, a procedure for selecting synthetic series and training a reinforcement algorithm with these data is developed.The Kolmogorov-Smirnov metric is used as an essential component to train GANs.The results of the experiments are explained, and the difficulty in calibrating generative adversarial and reinforcement network models is shown.Finally, conclusions derived from the project and possible future research are presented.