📄 中文摘要
本方案提出了一种利用变分量子电路（VQC）解决动态投资组合优化问题的量子强化学习（QRL）方法。该QRL实现借鉴了经典神经网络中深度确定性策略梯度（DDPG）和深度Q网络（DQN）算法的量子对应版本。通过在真实世界金融数据上的实证评估，我们的量子代理在风险调整后的性能方面展现出与经典方法相当的水平。具体而言，针对投资组合的动态管理，该方法将市场状态编码为量子态，并利用VQC作为策略网络或Q函数逼近器。在DDPG的量子版本中，VQC被用于参数化策略，直接输出在给定市场条件下对各类资产的投资权重。而在DQN的量子对应实现中，VQC则用于估计不同行动（如买入、卖出、持有特定资产）的Q值。训练过程通过量子线路的参数优化来实现，目标是最大化长期累积回报，同时考虑风险因素。实验结果表明，尽管量子硬件仍处于早期阶段，但这种基于VQC的QRL方法在处理复杂金融市场动态、实现有效资产配置方面具有潜力，为未来金融领域的量子计算应用奠定了基础。
📄 English Summary
Variational Quantum Circuit-Based Reinforcement Learning for Dynamic Portfolio Optimization
A Quantum Reinforcement Learning (QRL) solution to the dynamic portfolio optimization problem is presented, leveraging Variational Quantum Circuits (VQC). The implemented QRL approaches are quantum analogues of classical neural-network-based Deep Deterministic Policy Gradient (DDPG) and Deep Q-Network (DQN algorithms). Through an empirical evaluation on real-world financial data, the quantum agents achieve risk-adjusted performance comparable to classical methods. Specifically, for dynamic portfolio management, market states are encoded into quantum states, and VQCs serve as policy networks or Q-function approximators. In the quantum DDPG variant, VQCs parameterize the policy, directly outputting investment weights for various assets under given market conditions. For the quantum DQN counterpart, VQCs estimate Q-values for different actions, such as buying, selling, or holding specific assets. The training process involves optimizing the parameters of the quantum circuits to maximize long-term cumulative returns while considering risk factors. Experimental results indicate that, despite the early stage of quantum hardware, this VQC-based QRL approach demonstrates potential in handling complex financial market dynamics and achieving effective asset allocation, laying a foundation for future quantum computing applications in finance.