Real-Time Trading with Sequence Models and Reinforcement Learning
Introduction
Many claim that financial markets are unpredictable and inscrutable. That may be true, but it should not stop us from taking our best shot at deciphering this ultimate black box. This project aims to construct a novel framework for doing real-time trading on the foreign exchange and commodity markets. We will attempt this by leveraging two rapidly evolving fields of deep learning that are at the forefront of research and development: sequence models and deep reinforcement learning (DRL). We will construct the DRL environment from scratch, determining the observation space and how signals can be extracted from the observations. Furthermore, we will design the DRL agent itself to best interpret these inputs and make a trading decision. Time (and legally) permitting, we will attempt to connect the agent to a real broker and make some trades.
Ultimately, this project aims to be a valuable learning experience for those interested in reinforcement learning and finance. We are looking for developers who are first and foremost passionate and willing to dedicate themselves to the project, along with possessing basic machine learning skills. Experience in reinforcement learning, sequence models, and finance is a plus. If you are interested, email me at jingmin.wang@mail.utoronto.ca or aditya.srichandan@mail.utoronto.ca.
