Paper: arXiv 2509.12753

Authors: Feliks Bańka, Jarosław A. Chudziak

Abstract

In volatile financial markets, balancing risk and return remains a significant challenge. Traditional approaches often focus solely on equity allocation, overlooking the strategic advantages of options trading for dynamic risk hedging. This work presents DeltaHedge, a multi-agent framework that integrates options trading with AI-driven portfolio management. By combining advanced reinforcement learning techniques with an ensembled options-based hedging strategy, DeltaHedge enhances risk-adjusted returns and stabilizes portfolio performance across varying market conditions. Experimental results demonstrate that DeltaHedge outperforms traditional strategies and standalone models, underscoring its potential to transform practical portfolio management in complex financial environments. Building on these findings, this paper contributes to the fields of quantitative finance and AI-driven portfolio optimization by introducing a novel multi-agent system for integrating options trading strategies, addressing a gap in the existing literature.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • 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 leverages advanced deep reinforcement learning and ensemble methods, indicating significant mathematical density, while the experimental setup with benchmarks, ablation studies, and claims of outperformance suggests robust empirical validation.

Research Flowchart

  flowchart TD
  R["Research Goal: Enhance risk-adjusted returns via options"] --> I["Market Data Inputs<br>Historical Prices & Volatility"]
  I --> M["Key Methodology: Multi-Agent RL Framework<br>DeltaHedge with Ensembled Hedging Strategy"]
  M --> C["Computational Processes<br>Simulated Trading & Portfolio Optimization"]
  C --> O["Key Findings & Outcomes<br>Outperforms traditional strategies<br>Stabilizes portfolio across market conditions"]

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 DeltaHedge, a novel multi‑agent framework that integrates options trading with AI‑driven portfolio management.

Source sentence
this work presents DeltaHedge, a multi-agent framework that integrates options trading with AI-driven portfolio management.

Assumes 6

  • Protective options are near‑the‑money with a fixed 30‑day time‑to‑expiration, chosen for liquidity and pricing stability. market
    Source sentence
    Protective Options: Near-the-money options with a fixed time-to-expiration (e.g., 30 days), chosen based on liquidity and pricing stability.
  • Daily trading decisions are made on a daily time scale, processing new market data each step. execution
    Source sentence
    Our framework operates on a daily time scale, processing new market data at each step and adjusting the portfolio accordingly.
  • Transaction costs for equity trades are a linear function of price and trade size (0.2% per share). model
    Source sentence
    To account for trading expenses, we adopt a transaction cost model commonly used in prior research, given by: Ct = 0.002 × pt × |ΔNt|.
  • Option hedging costs include a fixed per‑contract fee and a 0.5% proportional fee on the premium. model
    Source sentence
    Ct hedge = nt ⋅0.007 + 0.005 ⋅nt ⋅Pt opt, where 0.007 is the fixed cost per contract (e.g., $0.70),
Show 2 more
  • Sufficient market volume and open interest exist to execute the desired option contracts; otherwise orders are scaled down. execution
    Source sentence
    if the available open interest or volume is below the required threshold, the order is deferred or scaled down proportionally.
  • Sentiment scores are derived from daily financial news headlines using a transformer‑based classifier. data
    Source sentence
    Sentiment Data: Daily sentiment scores computed from financial news headlines using a transformer-based sentiment classification model.

Data

Backtest sample: S&P 500 index · multi-asset · daily

  • Stock price data · Yahoo Finance market-prices public
    2010-01 to 2019-12 (S&P 500), 2016-01 to · S&P 500 index, Apple, Tesla
    Source sentence
    Stock price data: Daily closing prices and historical volatility from Yahoo Finance.
  • Options data · optionsdx.com options public
    Source sentence
    Options data: Retrieved from optionsdx.com, including key contract information used for hedging decisions.
  • Sentiment data · EOD Historical Data text-news commercial
    Source sentence
    Sentiment data: News headlines and content retrieved from EOD Historical Data, used to compute sentiment scores via a transformer-based classifier.
  • Volatility index (VIX) · Yahoo Finance macro public
    Source sentence
    Volatility index (VIX): Sourced from Yahoo Finance to capture overall market uncertainty.

Recognised sources: Yahoo Finance

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: Modeled - Equity trades: 0.2% of trade value (Ct = 0.002 × pt × |ΔNt|). Option trades: $0.70 fixed per contract plus 0.5% of premium (Ct hedge = nt·0.007 + 0.005·nt·Pt_opt).

Source sentence (the authors' words)
... Ct hedge = nt ⋅0.007 + 0.005 ⋅nt ⋅Pt opt, where 0.007 is the fixed cost per contract

Finds 5

Reported performance (DeltaHedge multi-agent options hedging framework): Sharpe 1.33 (benchmark 0.44) · total return 121.1% · max drawdown 10.1%

  • DeltaHedge outperforms traditional strategies and standalone RL models in risk‑adjusted returns. performance
    Source sentence
    Experimental results demonstrate that DeltaHedge outperforms traditional strategies and standalone models, underscoring its potential to transform practical portfolio management.
  • On Apple, the ensemble improves Sharpe by five basis points and halves maximum drawdown to 13%. performance
    Source sentence
    For Apple, the ensemble outperforms the best single hedger by only five basis points in Sharpe but halves the draw-down to 13 %.
  • On Tesla, the ensemble’s Sharpe advantage is small but reduces drawdown by three percentage points. performance
    Source sentence
    for Tesla, the Sharpe advantage is similarly small, yet draw-down falls by three percentage points.
  • Classic delta‑hedge is second‑best on Tesla but worst on Apple, showing the benefit of dynamic ensemble selection. performance
    Source sentence
    ... the classic delta hedge is second-best on Tesla but worst on Apple.
  • The ensemble‑based hedging strategy yields higher Sharpe and lower volatility compared to single‑hedger and no‑options variants. performance
    Source sentence
    By comparing our full ensemble strategy against the Single-Hedger and No-Options variants, we isolate exactly how dynamic, coordinated options hedging drives improvements in performance metrics.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 50 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive (1 by independent authors)

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 (42)

  1. Applying Informer for Option Pricing: A Transformer-Based Approach — → 2506.05565
  2. Large Language Models and the Elliott Wave Principle: A Multi-Agent Deep Learning Approach to Big Data Analysis in Financial Markets (2024) — link
  3. Toward Predictive Stock Trading with Hidformer Integrated into Reinforcement Learning Strategy (2024) — link
  4. Target-Oriented Multi-Agent Coordination with Hierarchical Reinforcement Learning (2024) — link
  5. Deep Learning for Options Trading: An End-To-End Approach — → 2407.21791
  6. Hedging Beyond the Mean: A Distributional Reinforcement Learning Perspective for Hedging Portfolios with Structured Products (2024) — link
  7. Dynamic asset allocation with asset-specific regime forecasts — → 2406.09578
  8. A Multimodal Foundation Agent for Financial Trading: Tool-Augmented, Diversified, and Generalist — → 2402.18485
and 34 more
  1. Combining transformer based deep reinforcement learning with Black-Litterman model for portfolio optimization — → 2402.16609
  2. Developing A Multi-Agent and Self-Adaptive Framework with Deep Reinforcement Learning for Dynamic Portfolio Risk Management — → 2402.00515
  3. Large Language Model based Multi-Agents: A Survey of Progress and Challenges (2024) — link
  4. Revolutionising Financial Portfolio Management: The Non-Stationary Transformer’s Fusion of Macroeconomic Indicators and Sentiment Analysis in a Deep Reinforcement Learning Framework (2023) — link
  5. Option pricing revisited: The role of price volatility and dynamics (2023) — link
  6. Implementing Portfolio Risk Management and Hedging in Practice — → 2309.15767
  7. TradingGPT: Multi-Agent System with Layered Memory and Distinct Characters for Enhanced Financial Trading Performance — → 2309.03736
  8. D-TIPO: Deep time-inconsistent portfolio optimization with stocks and options — → 2308.10556
  9. Option Return Predictability with Machine Learning and Big Data (2023) — link
  10. Systemic Risk of Optioned Portfolios: Controllability and Optimization (2022) — link
  11. A COMPREHENSIVE STUDY OF OPTION PRICING WITH TRANSACTION COSTS (2022) — link
  12. Transformers in Time Series: A Survey (2022) — link
  13. Derivatives-based portfolio decisions: an expected utility insight (2022) — link
  14. The Sharpe Ratio (2021) — link
  15. DeepTrader: A Deep Reinforcement Learning Approach for Risk-Return Balanced Portfolio Management with Market Conditions Embedding (2021) — link
  16. MSPM: A modularized and scalable multi-agent reinforcement learning-based system for financial portfolio management (2021) — link
  17. FinRL: A Deep Reinforcement Learning Library for Automated Stock Trading in Quantitative Finance (2020) — link
  18. Deep reinforcement learning for automated stock trading: an ensemble strategy — → 2511.12120
  19. Fast Agent-Based Simulation Framework with Applications to Reinforcement Learning and the Study of Trading Latency Effects (2020) — link
  20. Modelling Stock Markets by Multi-agent Reinforcement Learning (2020) — link
  21. MAPS: Multi-Agent reinforcement learning-based Portfolio management System (2020) — link
  22. Exploring Option Pricing and Hedging via Volatility Asymmetry (2020) — link
  23. Time-series forecasting with deep learning: a survey (2020) — link
  24. FinBERT: Financial Sentiment Analysis with Pre-trained Language Models (2019) — link
  25. Deep hedging (2018) — link
  26. Theory of Rational Option Pricing (2015) — link
  27. Long Short-Term Memory (1997) — link
  28. Option pricing: A simplified approach☆ (1979) — link
  29. Fact and Fantasy in the Use of Options (1975) — link
  30. Options Pricing Platform with Neural Networks, LLMs and Reinforcement Learning (2025) — link
  31. Unleashing the Power of Multi-Agent Reinforcement Learning for Algorithmic Trading in the Digital Financial Frontier and Enterprise Information Systems (2024) — link
  32. Enhanced Portfolio Optimization: Integrating Machine Learning and Risk Management Techniques (2024)
  33. Informer (2021)
  34. Options, Futures, and Other Derivatives (1989)

10 of 42 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.