基于深度强化学习的投资组合构建方法  

Portfolio construction method based on deep reinforcement learning

作　　者：李彬 潘乔[1] 阎希平 LI Bin;PAN Qiao;YAN Xiping(College of Computer Science and Technology,Donghua University,Shanghai 201620,China;Shanghai Zhaoqian Investment Co.LTD.,Shanghai 201107,China)

机构地区：[1]东华大学计算机科学与技术学院,上海201620 [2]上海兆前投资有限公司,上海201107

出　　处：《智能计算机与应用》2024年第8期85-90,共6页Intelligent Computer and Applications

摘　　要：传统基于数据分析的投资组合构建方法使用简单的统计学模型,不仅难以发现市场规律,且在处理大量数据时效率不高。而深度强化学习算法具备强大的数据处理和分析能力,能够通过学习自适应调整策略,从海量金融数据中提取出有效信息,处理复杂多变市场环境并为投资决策提供科学建议。针对金融资产价格具有非平稳特点和各资产间具有相互依赖性的问题,本文基于深度强化学习中的深度确定性策略梯度DDPG算法,设计了一种并行投资组合特征提取网络PPFNet作为策略网络用于构建投资组合。实验结果表明,PPFNet相较于其他主流投资组合构建方法,取得了最优的收益效益,且表现出良好的稳定性。Traditional portfolio construction methods based on data analysis often rely on simple statistical models that are unable to discover market patterns and can be inefficient when processing large amounts of data.In contrast,deep reinforcement learning algorithms possess powerful data processing and analysis capabilities,allowing them to extract valuable information from massive financial data,handle complex and changing market environments,and provide scientific advice for investment decisions by adapting strategies through self-learning.To address the issue of non-stationary asset price characteristics and interdependence between assets in the financial market,this paper proposes a parallel feature extraction network,PPFNet,based on the deep deterministic policy gradient(DDPG)algorithm in deep reinforcement learning as a policy network for constructing investment portfolios.Experimental results demonstrate that PPFNet outperforms other mainstream portfolio construction methods in terms of profit efficiency and exhibits excellent stability.

关 键 词：投资组合 深度强化学习 DDPG PPFNet 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]