FinHEAR: Human Expertise and Adaptive Risk-Aware Temporal Reasoning for Financial Decision-Making

Paper: arXiv 2506.09080

Authors: Jiaxiang Chen, Mingxi Zou, Zhuo Wang, Qifan Wang, Dongning Sun, Chi Zhang, Zenglin Xu

Abstract

Financial decision-making presents unique challenges for language models, demanding temporal reasoning, adaptive risk assessment, and responsiveness to dynamic events. While large language models (LLMs) show strong general reasoning capabilities, they often fail to capture behavioral patterns central to human financial decisions-such as expert reliance under information asymmetry, loss-averse sensitivity, and feedback-driven temporal adjustment. We propose FinHEAR, a multi-agent framework for Human Expertise and Adaptive Risk-aware reasoning. FinHEAR orchestrates specialized LLM-based agents to analyze historical trends, interpret current events, and retrieve expert-informed precedents within an event-centric pipeline. Grounded in behavioral economics, it incorporates expert-guided retrieval, confidence-adjusted position sizing, and outcome-based refinement to enhance interpretability and robustness. Empirical results on curated financial datasets show that FinHEAR consistently outperforms strong baselines across trend prediction and trading tasks, achieving higher accuracy and better risk-adjusted returns.

Complexity vs Empirical Score

• Math Complexity: 4.0/10
• Empirical Rigor: 6.0/10 (capped from 7.5: transaction costs only partially modelled; see penalty flags)
• Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper relies on LLM-based reasoning and agent coordination rather than advanced mathematical derivations, resulting in moderate math complexity. It demonstrates high empirical rigor with a curated dataset, public code release, and evaluation on multiple financial metrics (ACC, MCC, CR, SR, MDD, CalmarR).

Research Flowchart

  flowchart TD
  A["Research Goal: Improve Financial Decision-Making<br>with Adaptive Risk-Aware Reasoning"] --> B
  
  subgraph B ["Key Methodology: FinHEAR Multi-Agent Framework"]
      B1["Event-Centric<br>Pipeline"] --> B2["Historical Trend<br>Agent"]
      B1 --> B3["Current Event<br>Agent"]
      B1 --> B4["Expert Retrieval<br>Agent"]
      B1 --> B5["Confidence & Position<br>Sizing Agent"]
  end

  B --> C["Data & Inputs"]
  C --> C1["Historical Price Data"]
  C --> C2["Real-time Market Events"]
  C --> C3["Behavioral Economics<br>Knowledge Base"]
  C --> C4["Human Expert<br>Precedents"]

  B2 & B3 & B4 & B5 --> D["Computational Process"]
  D --> D1["Grounded Reasoning<br>Event Analysis"]
  D --> D2["Adaptive Risk Assessment<br>Loss Aversion & Feedback"]
  D --> D3["Dynamic Position Sizing<br>Confidence-Adjusted"]

  D1 & D2 & D3 --> E
  
  subgraph E ["Key Findings & Outcomes"]
      E1["Higher Trend Prediction<br>Accuracy"]
      E2["Better Risk-Adjusted<br>Returns"]
      E3["Enhanced Interpretability<br>& Robustness"]
  end

Cited References (75)

1. Long short-term memory (1997)
2. On the properties of neural machine translation: Encoder-decoder approaches (2014) — link
3. Attention is all you need (2017)
4. ALSTM: An attention-based long short-term memory framework for knowledge base reasoning (2020)
5. Proximal policy optimization algorithms (2017) — link
6. Informer: Beyond efficient transformer for long sequence time-series forecasting (2021)
7. A survey on deep learning for named entity recognition (2020)
8. Behavioral portfolio theory (2000)

1 of 75 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.