Paper: arXiv 2510.14988

Authors: Davide Ferrari, Alessandro Fulci, Sandra Paterlini

Abstract

Given a universe of N assets, investors often form equally weighted portfolios (EWPs) by selecting subsets of assets. EWPs are simple, robust, and competitive out-of-sample, yet the uncertainty about which subset truly performs best is largely ignored. Traditional approaches typically rely on a single selected portfolio, but this fails to consider alternative investment strategies that may perform just as well when accounting for statistical uncertainty. To address this selection uncertainty, we introduce the Selection Confidence Set (SCS) for EWPs: the set of all portfolios that, under a given loss function and at a specified confidence level, contains the unknown set of optimal portfolios under repeated sampling. The SCS quantifies selection uncertainty by identifying a range of plausible portfolios, challenging the idea of a uniquely optimal choice. Like a confidence set, its size reflects uncertainty – growing with noisy or limited data, and shrinking as the sample size increases. Theoretically, we establish that the SCS covers the unknown optimal selection with high probability and characterize how its size grows with underlying uncertainty, corroborating these results through Monte Carlo experiments. Applications to the French 17-Industry Portfolios and Layer-1 cryptocurrencies underscore the importance of accounting for selection uncertainty when comparing equally weighted strategies.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces a formal statistical framework with rigorous asymptotic coverage proofs and uses a Wald-type test, showing substantial mathematical depth, while also validating the theory with Monte Carlo simulations and applying it to real financial datasets (French industries and cryptocurrencies).

Research Flowchart

  flowchart TD
  A["Research Goal<br>Quantify Selection Uncertainty<br>in Equally Weighted Portfolios"] --> B["Methodology<br>Selection Confidence Set SCS"]
  
  B --> C["Data Inputs<br>French 17-Industry &<br>Layer-1 Crypto Assets"]
  
  C --> D{"Computation"}
  D --> E["Monte Carlo<br>Experiments"]
  D --> F["Empirical<br>Application"]
  
  E --> G["Theoretical Validation<br>Coverage Probabilities<br>& Size Growth"]
  F --> G
  
  G --> H["Key Findings<br>SCS Quantifies Selection Uncertainty<br>Size Grows with Noise<br>Validates Single-Portfolio Limitations"]
  
  style A fill:#e1f5fe,stroke:#01579b,stroke-width:2px
  style B fill:#fff3e0,stroke:#e65100,stroke-width:2px
  style C fill:#f3e5f5,stroke:#4a148c,stroke-width:2px
  style D fill:#fce4ec,stroke:#880e4f,stroke-width:2px
  style E fill:#e8f5e8,stroke:#2e7d32,stroke-width:2px
  style F fill:#e8f5e8,stroke:#2e7d32,stroke-width:2px
  style G fill:#fff9c4,stroke:#f57f17,stroke-width:2px
  style H fill:#ffebee,stroke:#c62828,stroke-width:2px

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 introduces the Selection Confidence Set (SCS) for equally weighted portfolios to quantify selection uncertainty.

Source sentence
we introduce the Selection Confidence Set (SCS) for EWPs: the set of all portfolios that, under a given loss function and at a specified confidence level, contains the unknown set of optimal portfolios

Assumes 6

  • The vector of estimators (µ̂s, σ̂2s, µ̂s′, σ̂2s′) jointly satisfies a central limit theorem. statistical
    Source sentence
    (A1) The vector of estimators (ˆµs, ˆσ2 s, ˆµs′, ˆσ2 s′) jointly satisfies the central limit theorem as in Equation (10), with asymptotic covariance matrix V(s, s′).
  • The estimated covariance matrix converges in probability to its population counterpart. statistical
    Source sentence
    (A2) Each element of the estimated covariance matrix ˆV(s, s′) converges in probability to its population counterpart ˆVjk(s, s′) p→Vjk(s, s′), for all j, k ∈{1, 2, 3, 4}.
  • Portfolio returns are i.i.d. across time for each selection. statistical
    Source sentence
    If portfolio returns Y (1) s , . . . , Y (T) s form an i.i.d. sequence for all s ∈S
  • The number of assets N is fixed and finite. model
    Source sentence
    assuming a fixed number of assets N < ∞
Show 2 more
  • The loss function L is continuously differentiable in mean and variance. model
    Source sentence
    We define the loss function as a continuously differentiable mapping L : R × R+ →R
  • The feasible selection set S may be restricted to reflect investment policies or transaction costs. other
    Source sentence
    While we focus on the general case where S includes all asset subset combinations, in practice restrictions may be introduced to reflect investment policies, transaction costs, or operational requirements.

Data

Backtest sample: L1-Cryptocurrency and 17 US industry portfolios · crypto · daily

  • L1-Crypto dataset · Yahoo Finance crypto-onchain public
    2022-01-01 to 2025-01-01 (T = 1095) · 16 Layer-1 cryptocurrencies
    Source sentence
    the L1-Crypto dataset, containing daily log-returns for 16 Layer-1 cryptocurrencies from January 1, 2022, to January 1, 2025 (T = 1095) (source: Yahoo Finance, https://finance.yahoo.com)
  • French 17 Industry dataset · K. French data library market-prices public
    2021-11-01 to 2024-10-31 (T = 755) · 17 equal-weighted U.S. industry portfolios
    Source sentence
    the French 17 Industry dataset, comprising daily returns for 17 equal-weighted U.S. industry portfolios from November 1, 2021, to October 31, 2024 (T = 755)

Costs

Unclear The classifier was not confident.

Finds 5

  • The Selection Confidence Set asymptotically contains the true optimal selection with at least the nominal confidence level. theoretical
    Source sentence
    the Selection Confidence Set bSα defined in (7) satisfies the asymptotic coverage property: lim T→∞P (S0 ⊆bSα) ≥1 −α.
  • When suboptimal models are well separated, the expected SCS size concentrates around |S0|(1‑α). theoretical
    Source sentence
    If γmin is bounded away from zero – i.e., all suboptimal models are well-separated from the optimal set – then the expected SCS size concentrates around |S0|(1 −α)
  • Monte Carlo shows the Ledoit‑Wolf bootstrap test yields larger SCS sizes than the Wald test (e.g., 447 vs 295 for T=100). performance
    Source sentence
    For example, with 1 −α = 0.95 and v = 1, the LW procedure yields an expected SCS size of about 447 for T = 100 and 43 for T = 1000, compared to 295 and 25, respectively, under the asymptotic approach.
  • In the L1‑Crypto data, the 95% SCS under Sharpe ratio contains 122 EWPs, versus 430 in the 17 Industry data. performance
    Source sentence
    at the 95% confidence level, the SCS under the Sharpe ratio for the L1-Cryptocurrency data includes 122 equally weighted portfolios (EWPs) that are statistically indistinguishable from an optimal EWP, compared to 430 in the 17 Industry dataset.
  • Relative Multiplicity Index is higher for Crypto (56.68) than Industry (48.54), indicating lower selection uncertainty for Crypto. statistical
    Source sentence
    the RMI index confirms that selection uncertainty is lower for the L1-Cryptocurrency data (RMI= 56.68) compared to that of the 17 Industry data (RMI= 48.54).

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 29 related papers). Coverage is partial, so this is neither replication nor refutation.

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 (148)

  1. The peer performance ratios of hedge funds (2018)
  2. Heteroskedasticity and autocorrelation consistent covariance matrix estimation (1991)
  3. poLCA : An R Package for Polytomous Variable Latent Class Analysis (2011) — link
  4. Applied latent class analysis (2002)
  5. Testing statistical hypotheses (2005)
  6. Connecting the dots: Econometric methods for uncovering networks with an application to the Australian financial institutions (2015)
  7. Partial correlation and conditional correlation as measures of conditional independence (2004)
  8. Model selection through sparse maximum likelihood estimation for multivariate Gaussian or binary data (2008)
and 140 more
  1. Conditional Sure Independence Screening (2016)
  2. New estimation approaches for graphical models with elastic net penalty (2022) — link
  3. The Trimmed Lasso: Sparsity and Robustness (2017) — link
  4. Convex optimization (2004)
  5. Algorithms for Minimization Without Derivatives (1973)
  6. Statistics for High-Dimensional Data: Methods, Theory and Applications (2011) — link
  7. Sustainable development and financial markets: old paths and new avenues (2015)
  8. A Constrained l1 Minimization Approach through sparse precision matrix estimation (2011)
  9. Valuing ESG: Doing Good or Sounding Good? (2020)
  10. Model selection in undirected graphical models with elastic-net (2011) — link
  11. Covariance Selection (1972)
  12. Variable selection and estimation with the seamless-L0 penalty (2013)
  13. Least angle regression (2004)
  14. Variable Selection via Nonconcave Penalized Likelihood and Its Oracle Properties (2001)
  15. Sure independence screening for ultrahigh dimensional feature space (2008)
  16. Sure independence screening in generalized linear models with NP-dimensionality (2010)
  17. Sparse inverse covariance estimation with the graphical lasso (2008)
  18. Extended Bayesian information criteria for Gaussian graphical models (2010)
  19. The Elements of Statistical Learning: Data Mining, Inference, and Prediction (2009) — link
  20. Statistical learning with sparsity: The Lasso and Generalizations (2016)
  21. Ridge regression: Biased estimation for nonorthogonal problems (1970)
  22. Probabilistic Graphical Models: Principles and Techniques (2009) — link
  23. Graphical elastic net and target matrices: Fast algorithm and software for sparse precision matrix estimation (2021) — link
  24. Precision matrix estimation with ROPE (2017)
  25. Sparse additive models (2007)
  26. Graphical Models (1996)
  27. Structure Learning of Gaussian Markov Random Fields with False Discovery Rate Control (2019)
  28. Variable selection via a combination of the l0 and l1 penalties (2007)
  29. SparseNet: Coordinate descent with nonconvex penalties (2011)
  30. Subset selection with shrinkage: Sparse linear modeling when SNR is low (2023)
  31. High-dimensional graphs and variable selection with the Lasso (2006)
  32. Subset selection in regression (2002)
  33. Machine Learning: A Probabilistic Perspective (2012)
  34. Gradient Method for Minimizing Composite functions (2013)
  35. A Simplex Algorithm for Function Minimization (1965)
  36. High-dimensional Ising model selection using $l_1$-regularized logistic regression (2010) — link
  37. Estimation of functional connectivity in fMRI data using stability selection-based sparse partial correlation with elastic net penalty (2012)
  38. Introduction to graphical modelling (2012)
  39. Variable selection via generalized SELO-penalized linear regression models (2018)
  40. A data driven Bayesian graphical ridge estimator (2022) — link
  41. Regression shrinkage and selection via the Lasso (1996)
  42. Robust and sparse banking network estimation (2018)
  43. Sparse graphical Gaussian modeling of the isoprenoid gene network in Arabidopsis thaliana (2004)
  44. Ridge estimation of inverse covariance matrices from high dimensional data (2016)
  45. Variable selection and parameter estimation with the Atan regularization method (2016) — link
  46. Collective dynamics of ’small-world’ networks (1998)
  47. High dimensional inverse covariance matrix estimation via linear programming (2010)
  48. Regularization and variable selection via elastic net (2005)
  49. The adaptive lasso and its oracle properties (2006)
  50. Modern machine learning solutions for portfolio selection (2021)
  51. Bayesian Mean–Variance Analysis: Optimal Portfolio Selection under Parameter Uncertainty (2021)
  52. Mean–Variance Portfolio Optimization Using Machine Learning-based Stock Price Prediction (2021)
  53. The Effect of Errors in Means, Variances, and Covariances on Optimal Portfolio Choice (2013)
  54. Optimal versus Naive Diversification: How Inefficient is the 1/N Portfolio Strategy? (2009)
  55. Size Matters: Optimal Calibration of Shrinkage Estimators for Portfolio Selection (2013)
  56. Multiple Hypothesis Testing in Microarray Experiments (2003)
  57. Risk and Volatility: Econometric Models and Financial Practice (2004)
  58. The Cross-Section of Expected Stock Returns (1992)
  59. Condence Sets for Model Selection by f-testing (2015)
  60. The Model Confidence set (2011)
  61. Deeply Equal-weighted Subset Portfolios (2020) — link
  62. Testing Statistical Hypotheses (2006)
  63. Equal-weighted strategy: Why it Outperforms Value-weighted Strategies? Theory and evidence (2017)
  64. Portfolio Selection (1952)
  65. The Markowitz Optimization Enigma: Is ‘optimized’ Optimal? (1989)
  66. Decision-making for Financial Trading: A Fusion Approach of Machine Learning and Portfolio Selection (2019)
  67. Why Do Equal-weighted Portfolios Outperform Value-Weighted Portfolios (2015)
  68. Exact and Approximate Stepdown Methods for Multiple Hypothesis Testing (2005)
  69. The Arbitrage Theory of Capital Asset Pricing (1976)
  70. Capital Asset Prices: A Theory of Market Equilibrium Under Conditions of Risk (1964)
  71. Portfolio Selection Problems with Markowitz’s Mean–variance Framework: a Review of Literature (2018)
  72. Model Selection Confidence Sets by Likelihood Ratio Testing (2019)
  73. Dependence function for continuous bivariate densities (1987)
  74. Graphical models for skew-normal variates (2003)
  75. Beyond normality: Learning sparse probabilistic graphical models in the non-Gaussian setting (2017) — link
  76. Esg, risk, and (tail) dependence (2023) — link
  77. Portfolio strategies and performance (1977)
  78. In defense of optimization: the fallacy of 1/n (2010)
  79. A well-conditioned estimator for large-dimensional covariance matrices (2004b)
  80. Asset allocation: combining investor views with market equilibrium (1990)
  81. Global portfolio optimization (1992)
  82. Honey, I shrunk the sample covariance matrix (2004a)
  83. Risk and asset allocation (2009)
  84. Condition-number-regularized covariance estimation (2013)
  85. Continuous Multivariate Distributions, Volume 1: Models and Applications (2004)
  86. A characterization of the distributions that imply mean-variance utility functions (1983)
  87. Graphical models via univariate exponential family distributions (2015)
  88. Simultaneous change point inference and structure recovery for high dimensional Gaussian graphical models (2021)
  89. Structure learning of undirected graphical models for count data (2021)
  90. The architecture of complex weighted networks (2004)
  91. Graphical lasso and thresholding: Equivalence and closed-form solutions (2019)
  92. Covariance regularization by thresholding (2008)
  93. huge: High-Dimensional Undirected Graph Estimation (2021) — link
  94. glasso: Graphical Lasso: Estimation of Gaussian Graphical Models (2019) — link
  95. GGMncv: Gaussian Graphical Models with Nonconvex Regularization (2021) — link
  96. foreach: Provides Foreach Looping Construct (2022) — link
  97. doParallel: Foreach Parallel Adaptor for the ‘parallel’ Package (2022) — link
  98. A 2-stage elastic net algorithm for estimation of sparse networks with heavy-tailed data (2023)
  99. On the Solution of $\ell_0$ -Constrained Sparse Inverse Covariance Estimation Problems (2021)
  100. A Risk Ratio Comparison of $\ell_0$ and $\ell_1$ Penalized Regressions (2010)
  101. $\ell_ 0 $ Sparse Inverse Covariance Estimation (2015)
  102. A novel $\ell_ 0 $-constrained gaussian graphical model for anomaly localization (2017)
  103. Estimation of a treatment effect based on a modified covariates method with L0 norm (2024)
  104. Long-term natural streamflow forecasting under drought scenarios using data-intelligence modeling (2024)
  105. Sparse HP filter: Finding kinks in the COVID-19 contact rate (2021)
  106. Sparsistency and rates of convergence in large covariance matrix estimation (2009)
  107. Variable selection and estimation with the seamless-L0 penalty (2013)
  108. Variable selection and estimation using a continuous approximation to the L0 penalty (2018)
  109. Model Selection in Gaussian Graphical Models: High-Dimensional Consistency of $\ell_1$ -Regularized MLE (2008)
  110. The Graphical Lasso: New Insights and Alternatives (2012)
  111. Variable Selection via Nonconcave Penalized Likelihood and Its Oracle Properties (2001) — link
  112. Stability Selection (2010) — link
  113. On Model Selection Consistency of Lasso (2006) — link
  114. Regularization and Variable Selection via the Elastic Net (2005) — link
  115. On the Solution of $\ell_0$-Constrained Sparse Inverse Covariance Estimation Problems (2021)
  116. Sparse Graphical Modelling via the Sorted $\ell_1$-Norm (2022) — link
  117. alabama: Constrained Nonlinear Optimization (2023) — link
  118. A note on the lack of symmetry in the graphical lasso (2013)
  119. On principal graphical models with application to gene network (2022) — link
  120. Robust performance hypothesis testing with the Sharpe ratio (2008)
  121. Robust performance hypothesis testing with smooth functions of population moments (2018)
  122. Robust performances hypothesis testing with the variance (2011)
  123. Source code for \texttt huge.generator in the \texttt huge R package
  124. MASS: Support Functions and Datasets for Venables and Ripley’s MASS (2009) — link
  125. Performance Hypothesis Testing with the Sharpe and Treynor Measures (1981) — link
  126. Performance Hypothesis Testing with the Sharpe Ratio (2003) — link
  127. Two is better than one: Regularized shrinkage of large minimum variance portfolios (2024)
  128. Bayesian portfolio selection using VaR and CVaR (2022)
  129. Fifty years of portfolio optimization (2024)
  130. Why Naive Diversification Is Not So Naive, and How to Beat It? (2024) — link
  131. S&P 500 Stock Selection Using Machine Learning Classifiers: A Look into the Changing Role of Factors (2024) — link
  132. Latent Structure Analysis (1968)
  133. Does Climate Policy Uncertainty Affect Chinese Stock Market Volatility? (2023) — link
  134. Climate Policy Uncertainty and World Renewable Energy Index Volatility Forecasting (2022) — link
  135. A Model Confidence Set Approach to the Combination of Multivariate Volatility Forecasts (2020) — link
  136. Improving Mean Variance Optimization through Sparse Hedging Restrictions (2015) — link
  137. High dimensional minimum variance portfolio estimation under statistical factor models (2021) — link
  138. Mixed‐frequency predictive regressions with parameter learning (2023) — link
  139. Incorporating Different Sources of Information for Bayesian Optimal Portfolio Selection (2025) — link
  140. Optimal Shrinkage-Based Portfolio Selection in High Dimensions (2022) — link

0 of 148 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.