Scholarlink

<p>Portfolio selection involves identifying the most promising stocks and determining their optimal proportions in a portfolio, balancing the trade-off between risk and return. This paper proposes a data-driven approach, based on Wasserstein distributionally robust optimization, for portfolio selection in the face of data uncertainty by combining asset pre-selection using a Support Vector Machine (SVM) machine learning model with portfolio optimization using the state-of-the-art conditional value-at-risk (CVaR) measure. A computationally efficient distributionally robust SVM classification model is formed, leveraging numerically tractable dual formulations of Wasserstein worst-case expectation problems, and solved using second-order cone programming techniques. Numerical experiments with real market data highlight that addressing distributional uncertainty in financial data during both stages of asset pre-selection and portfolio optimization results in superior cumulative portfolio returns.</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

Portfolio selection under data uncertainty: A blended distributionally robust approach of support vector machine and Mean-CVaR portfolio optimization

  • 期刊论文

  • 发表日期：

    2025年12月19日

查看全文
  • Neil D. Dizon,
  • Queenie Y. Huang,
  • Vaithilingam Jeyakumar

摘要

Portfolio selection involves identifying the most promising stocks and determining their optimal proportions in a portfolio, balancing the trade-off between risk and return. This paper proposes a data-driven approach, based on Wasserstein distributionally robust optimization, for portfolio selection in the face of data uncertainty by combining asset pre-selection using a Support Vector Machine (SVM) machine learning model with portfolio optimization using the state-of-the-art conditional value-at-risk (CVaR) measure. A computationally efficient distributionally robust SVM classification model is formed, leveraging numerically tractable dual formulations of Wasserstein worst-case expectation problems, and solved using second-order cone programming techniques. Numerical experiments with real market data highlight that addressing distributional uncertainty in financial data during both stages of asset pre-selection and portfolio optimization results in superior cumulative portfolio returns.