A Mean-VaR Based Deep Reinforcement Learning Framework for Practical Algorithmic Trading

It is difficult to automatically produce trading signals based on previous transaction data and the financial status of assets because significant noise unpredictability capital markets. This paper proposes an innovative algorithm solve optimal portfolio problem in stock market activities. Our novel strategy utilizes three features outperform other benchmark strategies a real-market environment. First, we propose mean-VaR optimization model, solution which actor-critic architecture. Unlike existing literature that learns expectation cumulative returns, critic module returns distribution by quantile regression, actor outputs weight maximizing objective function model. Secondly, use linear transformation realize short selling ensure investors have profit opportunities bear market. Third, A multi-process method, called Ape-x, was used accelerate speed deep reinforcement learning training. To validate our proposed approach, conduct backtesting for two representative portfolios observe model this work superior strategies.