Adaptive Portfolio Management in Volatile Markets Using Deep Reinforcement Learning

Asset prices are highly volatile as the market is driven by macro-economic events, sentiments and geopolitics, a lot of time is spent understanding the changing dynamics to be one step ahead in making investment and trading calls. Traditionally, these dynamic conditions have caused troubles to the traditional portfolio management system, which is based on static models, to allocate suboptimally. Such limitations are addressed in this research by using Deep Reinforcement Learning real time portfolio optimization earning (DRL). Specifically, we explore three DRL algorithms: Advantage Actor-Critic (A2C), Deep Deterministic Policy Gradient (DDPG), Proximal Policy Optimization(PPO). In this research, the DDPG outperforms the others in terms of its annual return 23.1%, the PPO has the highest Sharpe ratio 0.998 for risk adjusted returns and A2C achieves long-term stability with balanced risk.