An algorithm that can learn an optimal policy to execute trade profitable is any market participant’s dream. In the project, we propose an algorithm that does just that: a Deep Reinforcement Learning trading algorithm. We design our algorithm by tuning the reward function to our specified constraints, taking into account unrealized Profits and Losses (PnL), Sharpe ratio, profits, and transaction costs. Additionally, we use a short 5-month moving average replay memory in order to ensure our algorithm is basing its decision on the most pertinent information. We combine the aforementioned concepts to make a theoretical Deep Reinforcement Learning trading algorithm.
Bennett, Tucker; Ambrosen, Delaney; Woody, Joe; and Fruth, Simon, "Deep Reinforcement Learning in Trading Algorithms" (2018). IPHS 300: Artificial Intelligence for the Humanities: Text, Image, and Sound. Paper 9.