基于深度强化学习SAC算法的投资组合管理

Portfolio Management Based on Deep Reinforcement Learning SAC Algorithm

动态投资组合管理,是一种将资金按一定的计算规则,配置到不同金融股票产品中的决策过程。在中国市场上存在着大量股票以及金融产品,股票价格指数可以很好地反映中国的金融市场状况。SAC算法是基于最大熵的强化学习框架。在这个框架中,智能体在最大化期望奖励的同时,也最大化熵。将SAC算法运用于金融投资组合管理中,并与基准交易策略和先前的RL算法相比,获得17.53%的年回报率,证明此方法在盈利方面是可行的。

Dynamic portfolio management is a process of allocating funds to different financial stock products according to certain calculation rules.There are a large number of stocks and financial products in the Chinese market,and the stock index can well represent the financial mar ket situation in China.SAC(soft actor critic)algorithm is a Reinforcement Learning framework based on maximum entropy.In this frame work,the agent maximize the expected reward as well as the entropy.The SAC algorithm was applied to financial portfolio management and achieved an annual return of 17.53%compared with the benchmark trading strategy and the previous RL algorithm,proving that the meth od is profitable.