Scalp the Foreign Exchange Market with Deep Reinforcement Learning

This paper presents a reinforcement learning approach for foreign exchange trading. Inspired by technical analysis methods, this makes use of indicators encoding them into Gramian Angular Fields and searches patterns that indicate price movements using convolutional neural networks (CNN). In addition to the policy determines action take, an extra regression head is utilized determine size market orders. also experimentally shows maximizing return individual trade or cumulative reward in finite time window results better performance.