Paper: arXiv 2507.09916

Authors: Ahmad Aghapour, Erhan Bayraktar, Fengyi Yuan

Abstract

In this paper, we tackle the dynamic mean-variance portfolio selection problem in a {"\it model-free"} manner, based on (generative) diffusion models. We propose using data sampled from the real model $\mathbb P$ (which is unknown) with limited size to train a generative model $\mathbb Q$ (from which we can easily and adequately sample). With adaptive training and sampling methods that are tailor-made for time series data, we obtain quantification bounds between $\mathbb P$ and $\mathbb Q$ in terms of the adapted Wasserstein metric $\mathcal A W_2$. Importantly, the proposed adapted sampling method also facilitates {"\it conditional sampling"}. In the second part of this paper, we provide the stability of the mean-variance portfolio optimization problems in $\mathcal A W _2$. Then, combined with the error bounds and the stability result, we propose a policy gradient algorithm based on the generative environment, in which our innovative adapted sampling method provides approximate scenario generators. We illustrate the performance of our algorithm on both simulated and real data. For real data, the algorithm based on the generative environment produces portfolios that beat several important baselines, including the Markowitz portfolio, the equal weight (naive) portfolio, and S&P 500.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical machinery including score-based diffusion models, adapted Wasserstein metrics, and SDE stability analysis, while also providing empirical validation on both simulated and real financial data with performance comparisons against established baselines.

Research Flowchart

  flowchart TD
  A["Research Goal: Solve Dynamic Mean-Variance Portfolio Selection model-free"] --> B["Methodology: Score-based Diffusion Models & Policy Gradient"]
  B --> C["Data Inputs: Real Market Data (Limited Size)"]
  C --> D["Computation: Train Generative Model Q via Adapted Sampling"]
  D --> E["Computation: Approximate Scenario Generators & Optimize Portfolio"]
  E --> F["Outcomes: Error Bounds (Adapted Wasserstein) & Portfolio Stability"]
  F --> G["Findings: Algorithm beats Markowitz, Equal Weight, and S&P 500"]

Cited References (132)

  1. Approximating Mills ratio (2014) — link
  2. High-Dimensional Statistics: A Non-Asymptotic Viewpoint (2019) — link
  3. On logarithmic concave measures and functions (1973)
  4. Generating solution paths of Markovian stochastic differential equations using diffusion models (2025) — link
  5. Information Relaxations and Duality in Stochastic Dynamic Programs (2010) — link
  6. Information Relaxations, Duality, and Convex Stochastic Dynamic Programs (2014) — link
  7. A Financial Time Series Denoiser Based on Diffusion Models (2024) — link
  8. Generation of Synthetic Financial Time Series by Diffusion Models — → 2410.18897
and 124 more
  1. A Diffusion-Based Generative Model for Financial Time Series via Geometric Brownian Motion (2025) — link
  2. Classifier-Free Diffusion Guidance (2022) — link
  3. Unveil Conditional Diffusion Models with Classifier-Free Guidance: A Sharp Statistical Theory (2024) — link
  4. A Training-Free Conditional Diffusion Model for Learning Stochastic Dynamical Systems (2025) — link
  5. Data-Driven Hedging with Generative Models (2025) — link
  6. VolGAN : A Generative Model for Arbitrage-Free Implied Volatility Surfaces (2024)
  7. Diffusion Factor Models: Generating High-Dimensional Returns with Factor Structure (2025) — link
  8. Fitted Value Iteration Methods for Bicausal Optimal Transport — → 2306.12658
  9. Causal Transport in Discrete Time and Applications (2017)
  10. Numeraire-Invariant Quadratic Hedging and Mean–Variance Portfolio Allocation (2024)
  11. Convex duality in mean-variance hedging under convex trading constraints (2012)
  12. Convex Duality in Constrained Mean-Variance Portfolio Optimization (2007)
  13. Stochastic Stability of Differential Equations (2012)
  14. Recent advances in reinforcement learning in finance (2023)
  15. Adapted Wasserstein distances and stability in mathematical finance (2020) — link
  16. Sensitivity of Multiperiod Optimization Problems with Respect to the Adapted Wasserstein Distance (2023) — link
  17. Quantitative Fundamental Theorem of Asset Pricing (2025) — link
  18. Learning the Variance of the Reward-To-Go (2016) — link
  19. Accurate structure prediction of biomolecular interactions with AlphaFold 3 (2024)
  20. Equivariant Diffusion for Molecule Generation in 3 D (2022)
  21. Video Diffusion Models: A Survey (2024) — link
  22. High-resolution image synthesis with latent diffusion models (2022)
  23. Score-Based Generative Modeling through Stochastic Differential Equations (2021)
  24. Denoising Diffusion Probabilistic Models (2020)
  25. Generative modeling by estimating gradients of the data distribution (2019)
  26. Deep Unsupervised Learning using Nonequilibrium Thermodynamics (2015)
  27. Estimation of Non-Normalized Statistical Models by Score Matching (2005)
  28. A Simulation Approach to Dynamic Portfolio Choice with an Application to Learning About Return Predictability (2005)
  29. Consumption and Portfolio Decisions when Expected Returns are Time Varying (1999)
  30. Mean–Variance portfolio optimization with state-dependent risk aversion (2014)
  31. Machine Learning and Portfolio Optimization (2018)
  32. Empirical Asset Pricing via Machine Learning (2020)
  33. Trading strategies generated by Lyapunov functions (2017)
  34. Cover’s Universal Portfolio, Stochastic Portfolio Theory and the Numéraire Portfolio (2019)
  35. Universal Portfolios (1991)
  36. Dynamic Mean–Variance Asset Allocation (2010)
  37. Continuous-time mean–variance portfolio selection: A stochastic LQ framework (2000)
  38. Optimum Consumption and Portfolio Rules in a Continuous-Time Model (1971)
  39. Optimal Dynamic Portfolio Selection: Multiperiod Mean–Variance Formulation (2000)
  40. Optimal Multiperiod Portfolio Policies (1968)
  41. Portfolio Selection (1952)
  42. Score-Based Diffusion Models via Stochastic Differential Equations (2025) — link
  43. Addressing Function Approximation Error in Actor-Critic Methods (2018)
  44. Continuous Control with Deep Reinforcement Learning (2016)
  45. Mean–Variance Portfolio Selection by Continuous-Time Reinforcement Learning: Algorithms, Regret Analysis, and Empirical Study (2024) — link
  46. ScoreGrad : Multivariate Probabilistic Time Series Forecasting with Continuous Energy-based Generative Models (2021) — link
  47. Autoregressive denoising diffusion models for multivariate probabilistic time series forecasting (2021)
  48. Distribution Approximation and Statistical Estimation Guarantees of Generative Adversarial Networks (2022) — link
  49. Contractive Diffusion Probabilistic Models (2026) — link
  50. Score-based generative modeling secretly minimizes the Wasserstein distance (2022)
  51. Optimal Transport: Old and New (2008) — link
  52. Maximum likelihood training of score-based diffusion models (2021)
  53. Variance-Optimal Hedging in Discrete Time (1995) — link
  54. Diffusion models for time-series applications: a survey (2024)
  55. The mean-variance investment problem in a constrained financial market (2006) — link
  56. Time-consistent mean-variance portfolio selection in discrete and continuous time (2013) — link
  57. Distributionally robust mean-variance portfolio selection with Wasserstein distances (2022)
  58. Bounding adapted Wasserstein metrics (2024) — link
  59. Score approximation, estimation and distribution recovery of diffusion models on low-dimensional data (2023)
  60. Score-based generative models are provably robust: an uncertainty quantification perspective (2024) — link
  61. Sampling is as easy as learning the score: theory for diffusion models with minimal data assumptions (2023)
  62. Optimal versus naive diversification: How inefficient is the $1/N$ portfolio strategy? (2009)
  63. Machine learning portfolio allocation (2022)
  64. Deep learning with long short-term memory networks for financial market predictions (2018)
  65. Deep neural networks, gradient-boosted trees, random forests: Statistical arbitrage on the S&P 500 (2017)
  66. A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem (2017) — link
  67. AlphaPortfolio: Direct construction through deep reinforcement learning and interpretable AI (2021) — link
  68. Scenario reduction in stochastic programming (2003)
  69. An introduction to the bootstrap (1994)
  70. Monte Carlo methods in financial engineering (2004)
  71. Monte Carlo Methods for Value-at-Risk and Conditional Value-at-Risk: A Review (2014) — link
  72. Denoising diffusion probabilistic models (2020)
  73. Generating scenario trees for multistage decision problems (2001)
  74. Generative learning for financial time series with irregular and scale-invariant patterns (2024)
  75. Predict, refine, synthesize: Self-guiding diffusion models for probabilistic time series forecasting (2023)
  76. CorrGAN : Sampling realistic financial correlation matrices using generative adversarial networks (2020)
  77. The Markowitz optimization enigma: Is ‘optimized’optimal? (1989)
  78. Efficient Asset Management: A Practical Guide to Stock Portfolio Optimization and Asset Allocation. (2015) — link
  79. Sig-Wasserstein GANs for conditional time series generation (2024)
  80. Autoregressive denoising diffusion models for multivariate probabilistic time series forecasting (2021)
  81. Non-autoregressive conditional diffusion models for time series prediction (2023)
  82. CSDI: Conditional score-based diffusion models for probabilistic time series imputation (2021)
  83. Tail-GAN: Learning to Simulate Tail Risk Scenarios (2025) — link
  84. Quant GANs: deep generation of financial time series (2020)
  85. Time-series generative adversarial networks (2019)
  86. Scoregrad: Multivariate probabilistic time series forecasting with continuous energy-based generative models (2021) — link
  87. Which training methods for GANs do actually converge? (2018)
  88. Convergence of denoising diffusion models under the manifold hypothesis (2022)
  89. Wasserstein convergence guarantees for a general class of score-based generative models (2025)
  90. Long short-term memory (1997)
  91. U-Net : Convolutional networks for biomedical image segmentation (2015)
  92. Warrant Prices as Indicators of Expectations and Preferences (1961)
  93. Rational Theory of Warrant Pricing (1965)
  94. Risk Analysis in Capital Investment (1964)
  95. Option Pricing When Underlying Stock Returns Are Discontinuous (1976)
  96. A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options (1993)
  97. Jumps and Stochastic Volatility: Exchange‐Rate Processes Implicit in Deutsche Mark Options (1996)
  98. RiskMetrics—Technical Document (1996)
  99. Optimization of Conditional Value-at-Risk (2000)
  100. Efficient Asset Management: A Practical Guide to Stock Portfolio Optimization and Asset Allocation (1998)
  101. Monte Carlo Methods in Financial Engineering (2004)
  102. A non-elliptical orthogonal GARCH model for portfolio selection under transaction costs (2021)
  103. Portfolio Optimization Based on GARCH–EVT–Copula Forecasting Models (2018)
  104. International Asset Allocation with Regime Shifts (2002)
  105. On the Effectiveness of Scenario Generation Techniques in Single-Period Portfolio Optimization (2009)
  106. Quadratic Hedging and Mean-Variance Portfolio Selection with Random Parameters in an Incomplete Market (2004)
  107. Distributionally Robust Optimization under Moment Uncertainty with Application to Data-Driven Problems (2010)
  108. Data-Driven Distributionally Robust Optimization Using the Wasserstein Metric: Performance Guarantees and Tractable Reformulations (2018)
  109. Data-Driven Robust Optimization: A Comprehensive Survey (2019)
  110. Optimal Transport–Based Distributionally Robust Optimization: Structural Properties and Iterative Schemes (2022)
  111. Transaction-Cost–Aware Distributionally Robust Portfolio Optimization (2023)
  112. Modelling Extremal Events for Insurance and Finance (1997)
  113. Estimation of Tail-Related Risk Measures for Heteroscedastic Financial Time Series: An Extreme Value Approach (2000)
  114. Copula Methods in Finance (2004)
  115. Testing the Gaussian Copula Hypothesis for Financial Asset Dependences (2003)
  116. The Jackknife and the Bootstrap for General Stationary Observations (1989)
  117. The Stationary Bootstrap (1994)
  118. Block Bootstrap for Time Series: Convergence Rates and Tail Risks (2014)
  119. Generating Scenario Trees for Multistage Decision Problems (2001)
  120. Scenario Reduction in Stochastic Programming: An Approach Using Probability Metrics (2003)
  121. Denoising Diffusion Probabilistic Model for Realistic Financial Correlation Matrices (2024)
  122. Portfolio optimization using predictive auxiliary classifier generative adversarial networks (2023)
  123. Options: A monte carlo approach (1977)
  124. Stochastic Optimal Control: The Discrete-Time Case (1978)

2 of 132 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.