Algorithmic Trading of Stocks using Deep Reinforcement Learning
Machine Learning Developer | Deep Reinforcement Learning
This monograph project aims to explore the use of synthetic financial time series generated by a Generative Adversarial Network (GAN) model to train a Deep Q-Learning algorithm that executes buy and sell actions for a stock in the Standard & Poor's 500 index. To develop this project, I employed the CRISP DM methodology proposed by IBM, first understanding the business and necessary theory to develop the models, and then exploring and analyzing the available data that aligns with the study's objectives. This project develops a procedure for selecting synthetic series and training a reinforcement learning algorithm with this data. The Kolmogorov-Smirnov metric is used as an essential component to train the GAN networks. The results of the experiments are explained, and the difficulty of calibrating generative adversarial models and reinforcement learning agents is highlighted. Finally, the conclusions derived from the project and potential future research are presented.
Tech stack: Python · Keras · Financial Markets · RL · Deep Learning · AI · ML · GANs · Data Analysis · Algorithmic Trading · Synthetic Data