---
title: Robust enhanced index tracking portfolio selection under distributional uncertainty
url: https://www.ml-quant.com/papers/arxiv/2610.04221/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-10-09
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2610.04221
source_url: https://arxiv.org/abs/2610.04221
featured: 2026-10-09
citations: unknown
topic: Portfolio & Allocation
---


# Robust enhanced index tracking portfolio selection under distributional uncertainty

Robust optimization models minimize worst-case loss and downside risk for enhanced indexing under distributional uncertainty, beating benchmarks and standard robust approaches in out-of-sample wealth.

- Source: https://arxiv.org/abs/2610.04221
- Identifier: arXiv:2610.04221
- Released: 2026-10-06
- First featured: Quant Letter No. 134 (2026-10-09): https://www.ml-quant.com/issues/2026-10-09/
- Citations (Semantic Scholar): not tracked
- Published in: not yet
- Topic: Portfolio & Allocation
- Authors: Jun Cai, Zhiqiao Song

## Abstract (arXiv, CC0)

The enhanced index tracking (EIT) portfolio selection problem aims to construct a portfolio that is expected to outperform a benchmark index. In practice, investors face uncertainty in the joint distribution of asset and index losses, as the true distribution is typically unknown and only partial information is available. Moreover, portfolio selection requires balancing the trade-off between mean loss and downside risk while targeting higher returns. Effectively modeling and managing these factors in EIT portfolio selection is therefore of both practical and research interest. To address these challenges, this paper proposes robust EIT portfolio selection models under distributional uncertainty, guided by Pareto-optimality theory. The models minimize a convex combination of worst-case mean loss and downside risk, subject to a constraint on worst-case mean return. Two uncertainty sets are considered: one assuming a known mean loss vector and covariance matrix, and a more general set with unknown moments. Closed-form solutions are derived for the former, while the latter reduces to tractable convex optimization problems. Empirical results based on real market data show that the proposed models outperform not only the tracked index but also the equally weighted portfolio, robust mean-variance models, and robust variance models in terms of out-of-sample cumulative wealth, as well as Sharpe, Sortino, and Omega ratios. Moreover, portfolios that minimize a convex combination of worst-case mean loss and downside risk consistently outperform those that minimize downside risk alone. We also provide practical guidance for selecting optimal weighting coefficients to effectively balance risk and return.

## Related

- [The Critical Line Algorithm and the Constrained LASSO: One Curve, Two Literatures](https://www.ml-quant.com/papers/arxiv/2609.25704/): Shows that mean-variance portfolio selection and the constrained LASSO trace identical piecewise-linear solution paths, mapping their parametrizations exactly.
- [Signed network models for portfolio optimization](https://www.ml-quant.com/papers/arxiv/2510.05377/): The study shows that using negative edges in weighted signed network representations of financial markets can help reduce portfolio risk, performing on par with traditional models.
- [Bayesian Portfolio Optimization by Predictive Synthesis](https://www.ml-quant.com/papers/arxiv/2510.07180/): The article discusses a portfolio optimization technique using Bayesian predictive synthesis, which merges various asset return prediction models to handle financial market uncertainty.
- [Diffusion-Augmented Reinforcement Learning for Robust Portfolio Optimization under Stress Scenarios](https://www.ml-quant.com/papers/arxiv/2510.07099/): The research introduces a framework called DARL that combines DDPMs with DRL for portfolio management, improving its ability to withstand crises.
- [Adaptive and Regime-Aware RL for Portfolio Optimization](https://www.ml-quant.com/papers/arxiv/2509.14385/): The study presents a new reinforcement learning framework for portfolio optimization, which performs well under financial stress and supports dynamic asset allocation.
- [THEME: Enhancing Thematic Investing with Semantic Stock Representations and Temporal Dynamics](https://www.ml-quant.com/papers/arxiv/2508.16936/): Enhancing Thematic Investing: The study presents THEME, a hierarchical contrastive learning framework for thematic investing, which surpasses baselines in multiple retrieval metrics and enhances portfolio construction performance.