Application of A Deep Reinforcement Learning Method in Financial Market Trading

The analysis and design of financial trading systems (FTSs) is a topic of great interest both for the academic institutions and the professional one regarding to the promises by machine learning methodologies. In the paper, research is focus on the optimization of placement of limit order within a given time horizon and how to transpose the process into an end-to-end learning pipeline within the context of reinforcement learning. First, features were factored out and constructed from market data that related to movements of the Bitcoin/USD trading, which were used by deep reinforcement learning agents to learn a limit order placement policy then. A reinforcement learning environment was developed to emulate a local broker as well Finally, an evaluation procedure is defined to determine the capabilities and limitations of the policies learned by the reinforcement learning agents.