Deep reinforcement learning based multi-agent collaborated network for distributed stock trading

Abstract

Recently, interest in financial transactions is increasing, and the number of investors in the stock market is increasing. These investors are applying financial analysis methods to stock trading in order to gain more profits, and combining with artificial intelligence techniques has made it possible to achieve returns in excess of the market average. As a result, the stock trading system based on reinforcement learning has attracted attention, and in recent years, studies are being conducted to optimize financial time series data by Multi-Agent Reinforcement Learning (MARL). However, MARL, which is used in existing stock trading, cannot be fully collaborated because of lack of generalization of experience. Therefore, in this paper, we propose Multi-agent Collaborated Network (MCN) that can share and generalize the experience by agent, and experiment on collaboration in distributed stock trading.