Scholarlink

This paper shows how classic (Markowitz) and modern (CVaR) portfolio optimization models can be implemented efficiently in R. We compare two paradigms: the standard Sample Average Approximation (SAA) and distributionally robust optimization (DRO) using Wasserstein ambiguity sets. We first show that these advanced DRO models admit tractable reformulations as Second-Order Cone Programs (SOCPs). Then, we present the core contribution: a complete R pipeline that implements these exact SOCPs using the CVXR modeling package and the MOSEK solver. We provide code listings and discuss practical integration details. A numerical study on S&P 500 data illustrates SAA’s fragility, showing its consistent failure to meet out-of-sample return constraints. In contrast, the WDRO models successfully achieve this target, demonstrating a superior trade-off between performance and reliability. The result is a practical, reproducible toolkit for R users to build and solve advanced robust optimization problems.

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

From SAA to Distributionally Robust Portfolio Optimization in R: Markowitz and CVaR in Practice

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Diego Fonseca,
  • Mauricio Junca

摘要

This paper shows how classic (Markowitz) and modern (CVaR) portfolio optimization models can be implemented efficiently in R. We compare two paradigms: the standard Sample Average Approximation (SAA) and distributionally robust optimization (DRO) using Wasserstein ambiguity sets. We first show that these advanced DRO models admit tractable reformulations as Second-Order Cone Programs (SOCPs). Then, we present the core contribution: a complete R pipeline that implements these exact SOCPs using the CVXR modeling package and the MOSEK solver. We provide code listings and discuss practical integration details. A numerical study on S&P 500 data illustrates SAA’s fragility, showing its consistent failure to meet out-of-sample return constraints. In contrast, the WDRO models successfully achieve this target, demonstrating a superior trade-off between performance and reliability. The result is a practical, reproducible toolkit for R users to build and solve advanced robust optimization problems.