应用于投资组合优化的强化学习状态与值函数的选择

Reinforcement learning state and value function selection for portfolio optimization

针对金融资产未来收益的随机性,结合强化学习的原理,以Q-learning算法构造强化学习框架,来解决投资组合优化问题.采用一只股票连续数日开盘价和收盘价的涨跌幅信息作为状态,其中开盘信息在一定程度上纳入了市场消息面因素,收盘信息则直接反映股价波动情况,而连续数日的数据为模型加入了预测能力以用于指导投资.实证分析表明,投资者采用本研究方法进行投资可以得到较高且稳定的收益.

Aiming at the randomness of future returns of financial assets and the principle of reinforcement learning,this paper uses the Q-learning algorithm to construct a reinforcement learning framework to solve the problem of portfolio optimization.The article uses the information of the stock price’s opening and closing prices for several consecutive days as the status.The opening information has incorporated market news to a certain extent.The closing information directly reflects the stock price fluctuations,and the data for several consecutive days.Added predictive power to the model to guide investment.Empirical analysis shows that investors using this method to invest can get higher and stable returns.