Paper: arXiv 2502.13165

Abstract

As automated trading gains traction in the financial market, algorithmic investment strategies are increasingly prominent. While Large Language Models (LLMs) and Agent-based models exhibit promising potential in real-time market analysis and trading decisions, they still experience a significant -20% loss when confronted with rapid declines or frequent fluctuations, impeding their practical application. Hence, there is an imperative to explore a more robust and resilient framework. This paper introduces an innovative multi-agent system, HedgeAgents, aimed at bolstering system robustness via ``hedging’’ strategies. In this well-balanced system, an array of hedging agents has been tailored, where HedgeAgents consist of a central fund manager and multiple hedging experts specializing in various financial asset classes. These agents leverage LLMs’ cognitive capabilities to make decisions and coordinate through three types of conferences. Benefiting from the powerful understanding of LLMs, our HedgeAgents attained a 70% annualized return and a 400% total return over a period of 3 years. Moreover, we have observed with delight that HedgeAgents can even formulate investment experience comparable to those of human experts (https://hedgeagents.github.io/).

Complexity vs Empirical Score

  • Math Complexity: 2.0/10
  • Empirical Rigor: 6.0/10 (capped from 7.5: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper relies primarily on LLM agent architecture and system design rather than advanced mathematical modeling, but presents extensive empirical results with specific performance metrics (70% annualized return, 400% total return over 3 years), backtest comparisons against multiple baselines, and a live demo link.

Research Flowchart

  flowchart TD
  A["Research Goal: How to enhance LLM-based trading<br>robustness against market declines?"] --> B["Methodology: HedgeAgents Multi-Agent System<br>Central Manager + Hedging Experts"]
  B --> C["Inputs: Market Data &<br>Financial Asset Class Information"]
  C --> D["Computation: LLM Cognitive Reasoning &<br>Three Types of Agent Conferences"]
  D --> E["Process: Hedging Strategy Execution<br>& Real-time Coordination"]
  E --> F["Outcome: 70% Annualized Return<br>400% Total Return over 3 Years"]
  F --> G["Outcome: Formulated Investment Experience<br>Comparable to Human Experts"]

Cited References (206)

  1. Adversarial Video Moment Retrieval by Jointly Modeling Ranking and Localization (2020)
  2. Volatility as a transmitter of systemic risk: Is there a structural risk in finance? (2022)
  3. Deep learning volatility: a deep neural network perspective on pricing and calibration in (rough) volatility models (2021)
  4. Machine learning methods for systemic risk analysis in financial sectors. (2019)
  5. Pricing analysis of wind power derivatives for renewable energy risk management (2021)
  6. Quantitative methods in derivatives pricing: an introduction to computational finance (2003)
  7. Financial News Quantization and Stock Market Forecast Research Based on CNN and LSTM (2018) — link
  8. Deep Attentive Learning for Stock Movement Prediction From Social Media Text and Company Correlations (2020) — link
and 198 more
  1. Data fusion with factored quantization for stock trend prediction using neural networks (2023) — link
  2. Multi-Modal Relational Graph for Cross-Modal Video Moment Retrieval (2021)
  3. Keyword-Based Diverse Image Retrieval with Variational Multiple Instance Graph (2022)
  4. Point Prompt Tuning for Temporally Language Grounding (2022)
  5. Do LLMs Possess a Personality? Making the MBTI Test an Amazing Evaluation for Large Language Models (2023) — link
  6. Energy-based Automated Model Evaluation (2024) — link
  7. Contrastive topic-enhanced network for video captioning (2024)
  8. Adversarial cross-modal retrieval (2017)
  9. Cross-modal image-text retrieval with semantic consistency (2019)
  10. IMRAM: Iterative Matching With Recurrent Attention Memory for Cross-Modal Image-Text Retrieval (2020)
  11. A New Approach to Cross-Modal Multimedia Retrieval (2010)
  12. Modeling Information Retrieval by Formal Logic: A Survey (2014)
  13. From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions (2014)
  14. Retrieving Diverse Social Images at MediaEval 2016: Challenge, Dataset and Evaluation (2016)
  15. Div400: a social image retrieval result diversification dataset (2014)
  16. Diversifying Search Results using Self-Attention Network (2020)
  17. Diversified textual features based image retrieval (2019)
  18. Joint Image-Text Representation by Gaussian Visual-Semantic Embedding (2016)
  19. N- LTP : An Open-source Neural Language Technology Platform for Chinese (2021) — link
  20. A five-factor asset pricing model (2015) — link
  21. Stock Movement Prediction from Tweets and Historical Prices (2018) — link
  22. Listening to Chaotic Whispers (2018) — link
  23. SemEval 2015 Task 18: Broad-Coverage Semantic Dependency Parsing (2015) — link
  24. MRP 2020: The Second Shared Task on Cross-Framework and Cross-Lingual Meaning Representation Parsing (2020) — link
  25. Simpler but More Accurate Semantic Dependency Parsing (2018) — link
  26. Learning Joint Semantic Parsers from Disjoint Data (2018) — link
  27. Transition-Based Parsing for Deep Dependency Structures (2016) — link
  28. Lisbon: Evaluating TurboSemanticParser on Multiple Languages and Out-of-Domain Data (2015) — link
  29. Revisiting Pre-Trained Models for Chinese Natural Language Processing (2020) — link
  30. LERT: A Linguistically-motivated Pre-trained Language Model (2022) — link
  31. PERT: Pre-training BERT with Permuted Language Model (2022) — link
  32. Cognitive Endurance, Talent Selection, and the Labor Market Returns to Human Capital (2023) — link
  33. Democratization of Retail Trading: Can Reddit’s WallStreetBets Outperform Investment Bank Analysts? (2023) — link
  34. SKEP: Sentiment Knowledge Enhanced Pre-training for Sentiment Analysis (2020) — link
  35. Astock: A New Dataset and Automated Stock Trading based on Stock-specific News Analyzing Model (2022) — link
  36. Semantic dependency graph parsing using tree approximations (2015)
  37. Semantic role labeling: an introduction to the special issue (2008)
  38. Using EGARCH models to predict volatility in unconsolidated financial markets: the case of European carbon allowances (2023) — link
  39. An R-squared measure of goodness of fit for some common nonlinear regression models (1997) — link
  40. Weak Supervision for Fake News Detection via Reinforcement Learning (2020) — link
  41. PRUDEX-Compass: Towards Systematic Evaluation of Reinforcement Learning in Financial Markets (2023) — link
  42. Soft actor-critic algorithms and applications (2018) — link
  43. Training verifiers to solve math word problems (2021) — link
  44. Webgpt: Browser-assisted question-answering with human feedback (2021) — link
  45. Lamda: Language models for dialog applications (2022) — link
  46. Llama: Open and efficient foundation language models (2023) — link
  47. Proximal policy optimization algorithms (2017) — link
  48. Playing atari with deep reinforcement learning (2013) — link
  49. FinGPT: Open-Source Financial Large Language Models — → 2306.06031
  50. FinMem: A performance-enhanced LLM trading agent with layered memory and character design (2023) — link
  51. Efficient capital markets: A review of theory and empirical work (1970)
  52. Stock market volatility and learning (2016)
  53. Prune and tune ensembles: low-cost ensemble learning with sparse independent subnetworks (2022)
  54. No One Representation to Rule Them All: Overlapping Features of Training Methods (2022)
  55. An adaptive master-slave regularized model for unexpected revenue prediction enhanced with alternative data (2020)
  56. Computational intelligence and financial markets: A survey and future directions (2016)
  57. Uncertainty-aware lookahead factor models for quantitative investing (2020)
  58. Listening to chaotic whispers: A deep learning framework for news-oriented stock trend prediction (2018)
  59. The theory of the trading firm revisited (1978)
  60. Stock movement prediction from tweets and historical prices (2018)
  61. Investment behaviors can tell what inside: Exploring stock intrinsic properties for stock trend prediction (2019)
  62. Deep ensembles: A loss landscape perspective (2019) — link
  63. Stock selection via spatiotemporal hypergraph attention network: A learning to rank approach (2021)
  64. Stock market’s price movement prediction with LSTM neural networks (2017)
  65. CLVSA: A convolutional LSTM based variational sequence-to-sequence model with attention for predicting trends of financial markets (2019)
  66. Enhancing stock movement prediction with adversarial training (2018) — link
  67. Hierarchical multi-scale Gaussian transformer for stock movement prediction. (2020)
  68. Temporal relational ranking for stock prediction (2019)
  69. Deep attentive learning for stock movement prediction from social media text and company correlations (2020)
  70. Qlib: An AI-oriented quantitative investment platform (2020) — link
  71. Accurate multivariate stock movement prediction via data-axis transformer with multi-level contexts (2021)
  72. Learning multiple stock trading patterns with temporal routing adaptor and optimal transport (2021)
  73. The Sharpe ratio (1994)
  74. When quantitative trading meets machine learning: A pilot survey (2016)
  75. Algorithmic trading (2011)
  76. A comprehensive evaluation of ensemble learning for stock-market prediction (2020)
  77. Stock price prediction via discovering multi-frequency trading patterns (2017)
  78. Rest: Relational event-driven stock trend forecasting (2021)
  79. Long-tailed recognition by routing diverse distribution-aware experts (2020)
  80. Stock investment decision support using an ensemble of L-GEM based on RBFNN diverse trained from different years (2012)
  81. Predicting the stock market using multiple models (2006)
  82. Hybrid forecasting model research on stock data mining (2010)
  83. Bagging predictors (1996)
  84. DoubleEnsemble: A new ensemble method based on sample reweighting and feature selection for financial data analysis (2020)
  85. Random classification noise defeats all convex potential boosters (2010)
  86. Data arameters: A new family of parameters for learning a differentiable curriculum (2019)
  87. CatBoost: Unbiased boosting with categorical features (2018)
  88. Stock market forecast using multivariate analysis with bidirectional and stacked (LSTM, GRU) (2018)
  89. Multi-scale two-way deep neural network for stock trend prediction. (2020)
  90. Alphastock: A buying-winners-and-selling-losers investment strategy using interpretable deep reinforcement attention networks (2019)
  91. VolTAGE: Volatility forecasting via text-audio fusion with graph convolution networks for earnings calls (2020)
  92. Adaptive mixtures of local experts (1991)
  93. Learning factored representations in a deep mixture of experts (2013) — link
  94. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer (2017) — link
  95. Modeling task relationships in multi-task learning with multi-gate mixture-of-experts (2018)
  96. Scaling vision with sparse mixture of experts (2021)
  97. Deep mixture of experts via shallow embedding (2020)
  98. GLaM: Efficient scaling of language models with mixture-of-experts (2021) — link
  99. A dual-stage attention-based recurrent neural network for time series prediction (2017)
  100. Designing financial strategies based on artificial neural networks ensembles for stock markets (2018)
  101. Stock market index prediction using deep neural network ensemble (2017)
  102. Evaluating multiple classifiers for stock price direction prediction (2015)
  103. Performance of moving average trading strategies over varying stock market conditions: The Finnish evidence (2014)
  104. Deep learning with gated recurrent unit networks for financial sequence predictions (2018)
  105. A LSTM-based method for stock returns prediction: A case study of China stock market (2015)
  106. Stock market value prediction using neural networks (2010)
  107. Coherent measures of risk (1999)
  108. Conditional value-at-risk for general loss distributions (2002)
  109. Measuring systemic risk (2017)
  110. Measuring uncertainty (2015)
  111. Modeling financial uncertainty with multivariate temporal entropy-based curriculums (2021)
  112. Uncertainty in deep learning (2016)
  113. Selective Ensembles for Consistent Predictions (2021)
  114. BatchEnsemble: an Alternative Approach to Efficient Ensemble and Lifelong Learning (2019)
  115. Hyperparameter ensembles for robustness and uncertainty quantification (2020)
  116. Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time (2022) — link
  117. Film: Visual reasoning with a general conditioning layer (2018)
  118. Flipout: Efficient Pseudo-Independent Weight Perturbations on Mini-Batches (2018)
  119. Simple and scalable predictive uncertainty estimation using deep ensembles (2017)
  120. Snapshot ensembles: Train 1, get m for free (2017) — link
  121. Loss surfaces, mode connectivity, and fast ensembling of dnns (2018)
  122. Neural network ensembles (1990)
  123. Scalable bayesian optimization using deep neural networks (2015)
  124. A survey of uncertainty in deep neural networks (2021) — link
  125. Neural network ensembles, cross validation, and active learning (1994)
  126. Efficient and robust automated machine learning (2015)
  127. Hyperparameter optimization (2019)
  128. Popular ensemble methods: An empirical study (1999)
  129. Deep learning (2015)
  130. Ensemble selection from libraries of models (2004)
  131. What is being transferred in transfer learning? (2020)
  132. Averaging weights leads to wider optima and better generalization (2018)
  133. Cognitive architectures for language agents (2023) — link
  134. Generative Agents: Interactive simulacra of human behavior (2023) — link
  135. Toolformer: Language models can teach themselves to use tools (2023) — link
  136. Palm: Scaling language modeling with pathways (2023)
  137. Bloomberggpt: A large language model for finance (2023) — link
  138. Voyager: An open-ended embodied agent with large language models (2023)
  139. PRUDEX -Compass: Towards systematic evaluation of reinforcement learning in financial markets (2023)
  140. Reinforcement learning for quantitative trading (2023)
  141. TradeMaster: A holistic quantitative trading platform empowered by reinforcement learning (2023)
  142. Mastering stock markets with efficient mixture of diversified trading experts (2023)
  143. Random search for hyper-parameter optimization. (2012)
  144. Quantitative trading: how to build your own algorithmic trading business (2021)
  145. Deep reinforcement learning for quantitative trading: Challenges and opportunities (2022)
  146. FinRL : A deep reinforcement learning library for automated stock trading in quantitative finance (2020)
  147. Quantitative day trading from natural language using reinforcement learning (2021)
  148. Reinforcement-learning based portfolio management with augmented asset movement prediction states (2020)
  149. Investor-imitator: A framework for trading knowledge extraction (2018)
  150. DeepTrader: a deep reinforcement learning approach for risk-return balanced portfolio management with market conditions Embedding (2021)
  151. Commission fee is not enough: A hierarchical reinforced framework for portfolio management (2021)
  152. Earnhft: Efficient hierarchical reinforcement learning for high frequency trading — → 2309.12891
  153. DeepScalper: A risk-aware reinforcement learning framework to capture fleeting intraday trading opportunities (2022)
  154. Universal trading for order execution with oracle policy distillation (2021)
  155. Deep reinforcement learning for automated stock trading: An ensemble strategy (2020)
  156. Deep direct reinforcement learning for financial signal representation and trading (2016)
  157. Adaptive quantitative trading: An imitative deep reinforcement learning approach (2020)
  158. Auto-GPT for Online Decision Making: Benchmarks and Additional Opinions (2023) — link
  159. MetaGPT: Meta programming for a multi-agent collaborative framework (2023) — link
  160. AppAgent: Multimodal agents as smartphone users (2023) — link
  161. MM-REACT: Prompting ChatGPT for multimodal reasoning and action (2023) — link
  162. Chameleon: Plug-and-Play compositional reasoning with large language models (2023) — link
  163. GPT-4 Technical Report (2023) — link
  164. Chatgpt (2021) — [link](https://openai.com/ research/chatgpt)
  165. GPT-4V(ision) system card (2023) — link
  166. Pangu-Agent: A fine-tunable generalist agent with structured reasoning (2023) — link
  167. Chain-of-thought prompting elicits reasoning in large language models (2022)
  168. Why Can Large Language Models Generate Correct Chain-of-Thoughts? (2023) — link
  169. Talm: Tool augmented language models (2022) — link
  170. Learn to explain: Multimodal reasoning via thought chains for science question answering (2022)
  171. Dynamic prompt learning via policy gradient for semi-structured mathematical reasoning (2022) — link
  172. Theoremqa: A theorem-driven question answering dataset (2023) — link
  173. Scibench: Evaluating college-level scientific problem-solving abilities of large language models (2023) — link
  174. Scieval: A multi-level large language model evaluation benchmark for scientific research (2023) — link
  175. Mathvista: Evaluating mathematical reasoning of foundation models in visual contexts (2023) — link
  176. Generate rather than retrieve: Large language models are strong context generators (2022) — link
  177. Billion-scale similarity search with gpus (2019)
  178. O ptuna: A Next-Generation Hyperparameter Optimization Framework (2019)
  179. Technical analysis of stock trends (2018)
  180. Synapse: Trajectory-as-Exemplar Prompting with Memory for Computer Control (2023)
  181. Lightgbm: A highly efficient gradient boosting decision tree (2017)
  182. A survey on large language model based autonomous agents (2024) — link
  183. Investor sentiment and the mean–variance relation (2011) — link
  184. Dual Moments and Risk Attitudes (2018) — link
  185. Time series momentum (2012)
  186. DeepTrader: A Deep Reinforcement Learning Approach for Risk-Return Balanced Portfolio Management with Market Conditions Embedding (2021) — link
  187. A Multimodal Foundation Agent for Financial Trading: Tool-Augmented, Diversified, and Generalist (2024) — link
  188. FinReport: Explainable Stock Earnings Forecasting via News Factor Analyzing Model (2024) — link
  189. Glm-130b: An open bilingual pre-trained model (2022) — link
  190. Baichuan-13B (2023)
  191. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context (2024) — link
  192. Qwen-72B (2023)
  193. GPT-4 Technical Report (2024) — link
  194. GPT-3.5-Turbo: Advanced AI language model by OpenAI (2023) — link
  195. text-embedding-3-large (2023)
  196. Extreme market risk and extreme value theory (2013) — link
  197. Stock market volatility and economic policy uncertainty: New insight into a dynamic threshold mixed-frequency model (2024) — link
  198. FinReport: Explainable Stock Earnings Forecasting via News Factor Analyzing Model (2024) — link

2 of 206 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.