Scholarlink

<p>Portfolio optimization is a crucial decision-making component of intelligent financial system. It aims to maximize profits while managing risks by strategically allocating individual investments. Risk averter usually prefer the preservation of capital to the potential for a higher-than-average return, thus challenging existing portfolio optimization methods. This paper proposes a risk aware portfolio optimization method named RiskawareTrader for risk averter by incorporating downside risk, which could efficiently characterize losses. Specifically, this paper models the portfolio optimization as a Markov Decision Process, and adopts reinforcement learning method to get the optimum portfolio. RiskawareTrader adopts the downside risk in the reward function to penalize portfolios with high potential risks. In order to validate the performance of RiskawareTrader, this paper launches numerical experiments with some public datasets, namely the Dow Jones Industrial Average (DJIA) and the China Securities Index 300 (CSI300), in terms of annual return, maximum drawdown and Sharpe ratio. As demonstrated by the results, RiskawareTrader could achieve higher returns and better risk management under various market conditions compared to some baseline models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

RiskawareTrader: A Reinforcement Learning based Portfolio Optimization for Risk Averter

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月14日

查看全文
  • Min Yang,
  • Jin Wang,
  • Yi Hu

摘要

Portfolio optimization is a crucial decision-making component of intelligent financial system. It aims to maximize profits while managing risks by strategically allocating individual investments. Risk averter usually prefer the preservation of capital to the potential for a higher-than-average return, thus challenging existing portfolio optimization methods. This paper proposes a risk aware portfolio optimization method named RiskawareTrader for risk averter by incorporating downside risk, which could efficiently characterize losses. Specifically, this paper models the portfolio optimization as a Markov Decision Process, and adopts reinforcement learning method to get the optimum portfolio. RiskawareTrader adopts the downside risk in the reward function to penalize portfolios with high potential risks. In order to validate the performance of RiskawareTrader, this paper launches numerical experiments with some public datasets, namely the Dow Jones Industrial Average (DJIA) and the China Securities Index 300 (CSI300), in terms of annual return, maximum drawdown and Sharpe ratio. As demonstrated by the results, RiskawareTrader could achieve higher returns and better risk management under various market conditions compared to some baseline models.