Role of Reinforcement Learning in Financial Management Strategy

Algorithm trading is a hot topic in machine learning. Reinforcement learning is widely used in financial trading because it can directly teach behavioural rules using rational rewards. A typical financial trading strategy application is flexible and expresses the state of multiple stocks. Limited trading activities for financial products were possible due to the difficulty of designing a flexible action space. Due to the inherent transformation of the market base, it is difficult to extract effective characteristics during trading battles from price fluctuations in financial markets, including perhaps noise, and models learned from past price data. It does not work well with unknown price data. Thus, we achieved effective feature extraction from raw price data by DNN, online reinforcement learning adaptation to unknown price factors, and flexible asset management across multiple stocks in this study. The Casual dilated convolution layer is a DNN layer architecture that can capture long-term dependencies. The proposed method did not outperform the conventional method that applied online deep reinforcement learning to the portfolio management method of financial products, but 1x5 Convolution is the optimum filter size for this method's convolution layer.