基于条件生成式深度学习模型的大类资产配置策略研究

Research on Major Asset Allocation Strategies based on Conditional Generative Deep Learning Models

资产配置策略是金融投资领域中投资者们重点关注的一类问题,较为合理的资产配置策略不仅能够优化投资者的投资模式,而且能够为投资者们带来较为可观的超额收益。大类资产是投资者们在资产配置中常作选择的基底资产。基于此,本文拟采用生成式深度学习模型——GAN网络的衍生模型CGAN-GP网络对未来交易日中各项大类资产的投资风险进行追踪与预测,并选择传统资产配置模型——风险平价模型作为策略基本框架以设计智能资产配置策略,进行大类资产配置与量化交易。实证结果表明CGAN-GP网络所对应的策略净值具有较高的夏普比率与较高的年化收益率,且CGAN-GP网络具有较低的波动率,这也体现出CGAN-GP网络所对应策略的强稳健性。

Asset allocation strategy is a kind of problem that investors focus on in the field of financial investment.Reasonable asset allocation strategy can not only optimize investors’investment mode,but also bring investors considerable excess returns.Major assets are the base assets that investors often choose in asset allocation.Based on this,this paper intends to use the generative deep learning model—the derivative model of GAN network,CGAN-GP network,to track and forecast the investment risks of various types of assets in the future trading day,and choose the traditional asset allocation model—risk parity model as the basic framework of strategy to design intelligent asset allocation strategies,and conduct asset allocation and quantitative trading of broad categories.The empirical results show that the net worth strategy corresponding to the CGAN-GP network has a higher sharpe ratio and a higher annual return rate,and the CGAN-GP network has a low volatility,which also reflects the strong robustness of the strategy corresponding to the CGAN-GP network.