Deep Reinforcement Learning for Automated of Asian Stocks Trading

In a complex and changeable stock market, algorithmic stock trading has firmly established itself as a fundamental aspect of the present-day financial market, where most transactions are now fully automated. Additionally, Deep Reinforcement Learning (DRL) agents, renowned for their exceptional performance in intricate games such as chess and Go, are increasingly impacting the stock market. In this paper, we examine the potential of deep reinforcement learning to optimize the portfolio returns of 15 Asian stocks. We model stock trading as a Markov decision process problem because of its stochastic and interactive nature. Furthermore, we train a deep reinforcement learning agent using three actor-critic-based algorithms: proximal policy optimization (PPO), advantage actor-critic (A2C), and deep deterministic policy gradient (DDPG). We tested the algorithm on Asian stocks to see how well it performed pre-COVID, during COVID, and post-COVID. The trading agent’s performance using various reinforcement learning algorithms is assessed and compared to the traditional min-variance portfolio allocation strategy. The proposed three individual algorithms are above the minimum variance in risk-adjusted return as evaluated by the Sharpe ratio.