Paper: arXiv 2604.14206

Authors: Adhiraj Chattopadhyay

Abstract

This paper proposes a machine learning assisted portfolio optimization framework designed for low data environments and regime uncertainty. We construct a teacher student learning pipeline in which a Conditional Value at Risk (CVaR) optimizer generates supervisory labels, and neural models (Bayesian and deterministic) are trained using both real and synthetically augmented data. The synthetic data is generated using a factor based model with t copula residuals, enabling training beyond the limited real sample of 104 labeled observations. We evaluate four student models under a structured experimental framework comprising (i) controlled synthetic experiments (3 x 5 seed grid), (ii) in-distribution real market evaluation (C2A) and (iii) cross-universe generalization (D2A). In real-market settings, models are deployed using a rolling evaluation protocol where a frozen pretrained model is periodically fine tuned on recent observations and reset to its base state, ensuring stability while allowing limited adaptation. Results show that student models can match or outperform the CVaR teacher in several settings, while achieving improved robustness under regime shifts and reduced turnover. These findings suggest that hybrid optimization learning approaches can enhance portfolio construction in data constrained environments

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces an innovative semi-supervised teacher-student framework using factor-based t-copulas and Bayesian neural models to tackle label scarcity in portfolio optimization. It combines solid mathematical formulation with rigorous empirical evaluations across controlled synthetic environments, in-distribution markets, and cross-universe generalization tests.

Research Flowchart

  flowchart TD
    A["Research Goal: Portfolio Optimization under<br>Label Scarcity (104 obs) & Regime Shifts"] --> B["Data Inputs & Augmentation:<br>Real Market Data + Synthetic Factor Models (t-Copula)"]
    B --> C["Teacher Model:<br>Conditional Value at Risk (CVaR) Optimizer"]
    C -- "Generates Supervisory Labels" --> D["Semi-Supervised Sandwich Training:<br>Bayesian & Deterministic Student Neural Networks"]
    D --> E["Evaluation Protocol:<br>Synthetic Grid, In-Distribution (C2A) &<br>Cross-Universe (D2A) with Rolling Fine-Tuning"]
    E --> F["Key Outcomes:<br>Students Match/Outperform CVaR Teacher,<br>Lower Turnover & Enhanced Robustness"]

Cited References (38)

  1. Portfolio Selection (1952)
  2. Optimization of Conditional Value-at-Risk (2000)
  3. Global Portfolio Optimization (1992)
  4. Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk (1964)
  5. The Valuation of Risk Assets and the Selection of Risky Investments in Stock Portfolios and Capital Budgets (1965)
  6. Common Risk Factors in the Returns on Stocks and Bonds (1993)
  7. On Persistence in Mutual Fund Performance (1997)
  8. A Five-Factor Asset Pricing Model (2015)
and 30 more
  1. Learning to Trade via Direct Reinforcement (2001)
  2. Temporal Relational Ranking for Stock Prediction (2020)
  3. Stock Selection via Spatiotemporal Hypergraph Attention Network (2021)
  4. An Analysis of Variance Test for Normality (Complete Samples) (1965)
  5. Efficient Tests for Normality, Homoscedasticity and Serial Independence of Regression Residuals (1980)
  6. Asymptotic Theory of Certain “Goodness of Fit” Criteria Based on Stochastic Processes (1952)
  7. Sulla Determinazione Empirica di Una Legge di Distribuzione (1933)
  8. Optimization Proxies using Limited Labeled Data and Training Time – A Semi-Supervised Bayesian Neural Network Approach (2025) — link
  9. AlphaAgents: Large Language Model Based Multi-Agents for Equity Portfolio Constructions — → 2508.11152
  10. Empirical Asset Pricing via Machine Learning (2020) — link
  11. Machine Learning in a Data-Limited Regime: Augmenting Experiments with Synthetic Data Uncovers Order in Crumpled Sheets (2019) — link
  12. Weight Uncertainty in Neural Networks (2015) — link
  13. Adversarial Deep Reinforcement Learning in Portfolio Management (2018) — link
  14. Deep Portfolio Management: A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem (2017) — link
  15. Dynamic Copula Models and High Frequency Data (2014) — link
  16. Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning (2016) — link
  17. Deep Portfolio Theory (2018) — link
  18. A Review of Copula Models for Economic Time Series (2012) — link
  19. Dynamic Factor Copula Models with Estimated Cluster Assignments (2021) — link
  20. Deep Learning in Asset Pricing (2021) — link
  21. Deep Learning in Characteristics-Sorted Factor Models (2023) — link
  22. Deep Learning for Predicting Asset Returns (2018) — link
  23. Deep Partial Least Squares for Empirical Asset Pricing (2022) — link
  24. Deep Reinforcement Learning for Optimal Asset Allocation Using DDPG with TiDE — → 2508.20103
  25. Asset Pricing and Machine Learning: A critical review (2024) — link
  26. Four Factor Model in Indian Equities Market (2014)
  27. A Mathematical Theory of Communication (1948) — link
  28. Auto-Encoding Variational Bayes (2013) — link
  29. Ridge Regression: Biased Estimation for Nonorthogonal Problems (1970) — link
  30. yfinance: Download market data from Yahoo! Finance’s API (2019) — link

2 of 38 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.