Paper: arXiv 2611.14699

Authors: Mohammad Zarei, Amirhossein MohammadpourAzari, Moslem Habibi

Abstract

Reinforcement-learning portfolio policies can optimize allocations from prices, covariances, and prior positions, yet they lack a calibrated, replayable mechanism for certifying that a proposed allocation respects a bounded risk envelope before execution. Regulators increasingly expect automated portfolio systems to provide exactly this kind of pre-trade audit trail, interpretable risk limit, and human-oversight mechanism. Gated Conformal Semantic Risk Control addresses this gap by separating return-seeking allocation from text-conditioned downside-risk verification. An instruction-tuned language model maps timestamped disclosures into schema-validated semantic risk states over macro, liquidity, earnings, and sector channels. These states induce correlated ellipsoidal factor-shock uncertainty sets whose conformal radii are calibrated by rolling finite-sample prediction over localized semantic nonconformity buckets. An asymmetric second-order cone gate then checks each raw PPO proposal against a certified worst-case semantic loss budget. The gate stays inactive for compliant proposals and intervenes only when that budget is exceeded. In walk-forward out-of-sample tests on nine liquid U.S. Select Sector SPDR ETFs, Raw PPO earns a 12.7% annual return while incurring a 4.05% semantic violation rate. Gated-CSRC removes all observed out-of-sample semantic limit breaches while retaining an 11.7% annual return. This 100-basis-point auditable safety premium is smaller than the yield loss imposed by unconditional hard projection, and exact McNemar testing confirms the compliance gain at $p < 10^{-15}$. Loughran-McDonald lexicon controls and time-shuffled text controls fail to reproduce this certificate.

Complexity vs Empirical Score

  • Math Complexity: 8.0/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: This paper presents a highly novel approach combining reinforcement learning with conformal prediction and NLP for risk control. The empirical validation is robust, demonstrating a significant compliance gain with a minimal return trade-off. The mathematical underpinnings of conformal prediction and robust optimization are sophisticated.

Research Flowchart

  flowchart TD
    A[Research Goal: Certify RL Portfolio Allocations Under Bounded Risk] --> B{Methodology: Gated Conformal Semantic Risk Control (G-CSRC)};
    B --> C[Inputs: Prices, Covariances, Prior Positions, Timestamped Disclosures (Text)];
    C --> D{Computational Process 1: Text-Conditioned Semantic Risk States via LLM};
    D --> E{Computational Process 2: Conformal Prediction for Uncertainty Set Calibration};
    E --> F{Computational Process 3: Asymmetric Second-Order Cone Gate (SOCG) on PPO Proposals};
    F --> G[Outcome: G-CSRC eliminates semantic breaches (100bps return trade-off) with statistical significance];

Cited References (52)

  1. Regulation ( EU ) 2024/1689 of the European Parliament and of the Council of 13 June 2024 Laying Down Harmonised Rules on Artificial Intelligence (2024) — link
  2. Use of Derivatives by Registered Investment Companies and Business Development Companies (2020) — link
  3. Portfolio Selection (1952) — link
  4. Optimization of conditional value-at-risk (2000)
  5. Learning to trade via direct reinforcement (2001)
  6. A deep reinforcement learning framework for the financial portfolio management problem (2017) — link
  7. Cost-Sensitive Portfolio Selection via Deep Reinforcement Learning (2022) — link
  8. Sentiment Trading with Large Language Models — → 2412.19245
and 44 more
  1. From Text to Alpha: Can LLMs Track Evolving Signals in Corporate Disclosures? (2026) — link
  2. Large Language Models as Financial Analysts: Sector-Aware Reasoning for Investment Decisions (2026) — link
  3. Algorithmic Learning in a Random World (2005) — link
  4. Risk-Sensitive Policies for Portfolio Management (2022) — link
  5. Twin-System Recurrent Reinforcement Learning for Optimizing Portfolio Strategy (2024) — link
  6. Portfolio Construction Using Explainable Reinforcement Learning (2024) — link
  7. Large Language Models in Equity Markets: Applications, Techniques, and Insights (2025) — link
  8. Robust Convex Optimization (1998) — link
  9. Robust Portfolio Selection Problems (2003) — link
  10. The Price of Robustness (2004) — link
  11. Distributionally Robust Optimization under Moment Uncertainty with Application to Data-Driven Problems (2010) — link
  12. Deep Graph Convolutional Reinforcement Learning for Financial Portfolio Management (2021) — link
  13. WaveCorr : Deep Reinforcement Learning with Permutation Invariant Convolutional Policy Networks for Portfolio Management (2023) — link
  14. Time series is not enough: Financial Transformer Reinforcement Learning for portfolio management (2025)
  15. An Adaptive Portfolio Trading System: A Risk-Return Portfolio Optimization Using Recurrent Reinforcement Learning with Expected Maximum Drawdown (2017) — link
  16. A Constrained Portfolio Trading System Using Particle Swarm Algorithm and Recurrent Reinforcement Learning (2019) — link
  17. Portfolio Constructions in Cryptocurrency Market: A CVaR -Based Deep Reinforcement Learning Approach (2023) — link
  18. Constrained Policy Optimization (2017) — link
  19. When Is a Liability Not a Liability? T extual Analysis, Dictionaries, and 10-Ks (2011) — link
  20. Giving Content to Investor Sentiment: The Role of Media in the Stock Market (2007) — link
  21. FinBERT : Financial Sentiment Analysis with Pre-trained Language Models (2019) — link
  22. FinGPT : Open-Source Financial Large Language Models — → 2306.06031
  23. Adaptive and explainable margin trading via large language models on portfolio management (2024)
  24. GEMs-LLM : Integrating Large Language Models with Goal-Aware Exploration for RL -Based Portfolio Optimization (2025)
  25. On Calibration of Modern Neural Networks (2017) — link
  26. `` W hy Should I Trust You?’’: Explaining the Predictions of Any Classifier (2016) — link
  27. A unified approach to interpreting model predictions (2017)
  28. Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead (2019) — link
  29. Algorithmic Learning in a Random World (2005) — link
  30. A Tutorial on Conformal Prediction (2008) — link
  31. A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification (2021) — link
  32. Global Financial Stability Report, October 2024: Steadying the Course: Uncertainty, Artificial Intelligence, and Financial Stability (2024) — link
  33. Supervisory Guidance on Model Risk Management (2026) — link
  34. Proximal Policy Optimization Algorithms (2017) — link
  35. FNSPID : A Comprehensive Financial News Dataset in Time Series — → 2402.06698
  36. The Llama 3 Herd of Models (2024) — link
  37. Transformers: State-of-the-Art Natural Language Processing (2020) — link
  38. QLoRA : Efficient Finetuning of Quantized LLMs (2023)
  39. Optimal versus Naive Diversification: How Inefficient Is the 1/N Portfolio Strategy? (2009) — link
  40. Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency (1993) — link
  41. Note on the Sampling Error of the Difference between Correlated Proportions or Percentages (1947) — link
  42. Performance Hypothesis Testing with the Sharpe and Treynor Measures (1981) — link
  43. Performance Hypothesis Testing with the Sharpe Ratio (2003)
  44. A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix (1987) — link

3 of 52 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2570 linked papers.