Paper: arXiv 2508.10776

Authors: Juchan Kim, Inwoo Tae, Yongjae Lee

Abstract

Portfolio optimization constitutes a cornerstone of risk management by quantifying the risk-return trade-off. Since it inherently depends on accurate parameter estimation under conditions of future uncertainty, the selection of appropriate input parameters is critical for effective portfolio construction. However, most conventional statistical estimators and machine learning algorithms determine these parameters by minimizing mean-squared error (MSE), a criterion that can yield suboptimal investment decisions. In this paper, we adopt decision-focused learning (DFL) - an approach that directly optimizes decision quality rather than prediction error such as MSE - to derive the global minimum-variance portfolio (GMVP). Specifically, we theoretically derive the gradient of decision loss using the analytic solution of GMVP and its properties regarding the principal components of itself. Through extensive empirical evaluation, we show that prediction-focused estimation methods may fail to produce optimal allocations in practice, whereas DFL-based methods consistently deliver superior decision performance. Furthermore, we provide a comprehensive analysis of DFL’s mechanism in GMVP construction, focusing on its volatility reduction capability, decision-driving features, and estimation characteristics.

Complexity vs Empirical Score

  • Math Complexity: 9.0/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)

Why this score: The paper is highly mathematically dense, featuring complex theoretical derivations and proofs, but also demonstrates strong empirical rigor with extensive experiments and a detailed, end-to-end implementation framework for decision-focused learning in portfolio optimization.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Optimize GMVP using<br>Decision-Focused Learning"] --> B["Data & Inputs:<br>Historical Equity Returns<br>Asset Covariance Matrix"]
  B --> C["Methodology:<br>Apply DFL to directly<br>optimize portfolio decisions<br>vs. traditional MSE methods"]
  C --> D["Computational Process:<br>Derive analytic gradient<br>of GMVP loss function<br>via Principal Components"]
  D --> E["Key Outcomes:<br>DFL reduces volatility &<br>outperforms MSE-based<br>estimators in decision quality"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a decision-focused learning framework for constructing the global minimum-variance portfolio.

Source sentence
we adopt decision-focused learning (DFL)—an approach that directly optimizes decision quality rather than prediction error such as MSE—to derive the global minimum-variance portfolio (GMVP).

Assumes 3

  • Return stationarity is assumed when calibrating MVO parameters in‑sample. market
    Source sentence
    MVO parameters are typically calibrated using an in‑sample period under the assumption of return stationarity
  • Kernels of (Σ⁻¹)² and (wwᵀ)⊗(Σ⁻¹)² are non‑trivial. model
    Source sentence
    We assume that the kernels of (Σ−1)2 and (𝑤𝑤⊤) ⊗ (Σ−1)2 are non-trivial.
  • The GMVP is unconstrained, allowing a closed‑form solution. model
    Source sentence
    We focus on the unconstrained global minimum-variance portfolio (GMVP), which admits a closed-form solution

Data

Backtest sample: 2010-01 – 2024-12 · S&P 100, Industry, Dow Jones stocks · equities · daily

  • 49 Industry · Kenneth R. French data library market-prices public
    2010-01-01 to 2024-12-31 · 49 industry portfolios
    Source sentence
    First, we employ the 49 industry portfolios from the Kenneth R. French data library1, hereafter denoted “49 Industry.”
  • S&P 100 · S&P 500 constituents (market cap selection) market-prices public
    2010-01-01 to 2024-12-31 · top 100 S&P 500 stocks by market cap as of April 1, 2025
    Source sentence
    Second, we select the top 100 S&P 500 stocks by market capitalization as of April 1, 2025 (denoted “S&P 100”).
  • Dow 30 market-prices public
    2010-01-01 to 2024-12-31 · Dow Jones 30 index constituents
    Source sentence
    Third, we use the constituents of the Dow Jones 30 index (“Dow 30”).

Recognised sources: Kenneth French Data Library

Costs

Unclear - sources disagree The platform classifier and the paper-text extraction give different answers, so no value is asserted.

Platform classifier: Not modeled

Paper text extraction: Not applicable

Finds 3

  • DFL‑based methods consistently deliver superior decision performance compared with MSE‑based (PFL) models and the equally weighted benchmark. performance
    Source sentence
    DFL‑based methods consistently deliver superior decision performance.
  • DFL achieves higher precision in selecting low‑volatility stocks for long positions than OAS and PFL models. performance
    Source sentence
    DFL achieves high precision for long‑position selection based on low historical volatility.
  • DFL’s portfolio volatility increases with longer forecast horizons (larger δout), but is robust to the length of the input window (δin). performance
    Source sentence
    As δout increases for a fixed δin, DFL’s volatility rises, reflecting the greater uncertainty in longer‑horizon forecasts. Conversely, for a fixed δout, performance varies only modestly with δin, indicating robustness to the length of the input window.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 31 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 3 papers in the archive (3 by independent authors): Decision-Induced Ranking Explains Prediction Inflation and … (uses its method or data)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (35)

  1. Portfolio selection (1952)
  2. The effect of estimation risk on optimal portfolio choice (1976)
  3. Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy? (2009)
  4. Minimum-variance portfolio composition (2011)
  5. Minimum-variance portfolios in the US equity market (2006)
  6. Non-linear shrinkage of the price return covariance matrix is far from optimal for portfolio optimization (2023)
  7. On estimating the expected return on the market: An exploratory investigation (1980)
  8. On the sensitivity of mean-variance-efficient portfolios to changes in asset means: some analytical and computational results (1991)
and 27 more
  1. The effect of errors in means, variances and covariances on optimal portfolio choice (1993)
  2. Dominating estimators for minimum-variance portfolios (2010)
  3. Deep learning for volatility forecasting in asset management (2022)
  4. Decision-Aware Learning for Optimizing Health Supply Chains (2022) — link
  5. Anatomy of Machines for Markowitz: Decision-Focused Learning for Mean-Variance Portfolio Optimization — → 2409.09684
  6. Smart “Predict, then Optimize” (2022)
  7. Decision-Focused Learning: Through the Lens of Learning to Rank (2022)
  8. Improved estimation of the covariance matrix of stock returns with an application to portfolio selection (2003)
  9. Nonlinear Shrinkage Estimation of Large‐Dimensional Covariance Matrices (2012)
  10. Empirical Bayes Estimation of the Multivariate Normal Covariance Matrix (1980)
  11. Bayesian estimation of the global minimum variance portfolio (2017)
  12. Estimation of the global minimum variance portfolio in high dimensions (2018)
  13. Cross validation based transfer learning for cross-sectional non-linear shrinkage: A data-driven approach in portfolio optimization (2024)
  14. Decision-Focused Learning: Foundations, State of the Art, Benchmark and Future Opportunities (2024)
  15. Predicting Wildlife Trafficking Routes with Differentiable Shortest Paths (2023)
  16. Port Selection for Fluid Antenna Systems (2022)
  17. OptNet: Differentiable Optimization as a Layer in Neural Networks (2017)
  18. Differentiable Convex Optimization Layers (2019)
  19. Decision‐Focused Learning without Decision‐Making: Learning Locally Optimized Decision Losses (2022)
  20. Integrating Prediction in Mean–Variance Portfolio Optimization (2023)
  21. Distributionally Robust End‑to‑End Portfolio Construction (2023)
  22. Are Transformers Effective for Time Series Forecasting? (2023)
  23. Analytical Nonlinear Shrinkage of Large-Dimensional Covariance Matrices (2020)
  24. A Well‑Conditioned Estimator for Large‑Dimensional Covariance Matrices (2004)
  25. Shrinkage Algorithms for MMSE Covariance Estimation (2010)
  26. Adam: A Method for Stochastic Optimization (2014) — link
  27. An Overview of Machine Learning for Portfolio Optimization. (2024)

1 of 35 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.