Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

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

TL;DR

FinHEAR tackles the challenge of applying large language models to financial decision-making by introducing a six-agent architecture that grounds predictions in structured human expertise and adaptive risk assessment within an event-driven pipeline. It embeds expert-guided retrieval, risk-aware position sizing guided by Prospect Theory, and a feedback-driven temporal refinement loop to maintain consistency with realized outcomes. Empirical results on trend forecasting and trading tasks across multiple assets show FinHEAR outperforms strong baselines in both accuracy and risk-adjusted returns, with ablations confirming the value of each component. The framework advances interpretable, robust, and human-aligned decision support for real-world financial environments.

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.

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

TL;DR

FinHEAR tackles the challenge of applying large language models to financial decision-making by introducing a six-agent architecture that grounds predictions in structured human expertise and adaptive risk assessment within an event-driven pipeline. It embeds expert-guided retrieval, risk-aware position sizing guided by Prospect Theory, and a feedback-driven temporal refinement loop to maintain consistency with realized outcomes. Empirical results on trend forecasting and trading tasks across multiple assets show FinHEAR outperforms strong baselines in both accuracy and risk-adjusted returns, with ablations confirming the value of each component. The framework advances interpretable, robust, and human-aligned decision support for real-world financial environments.

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.

Paper Structure

This paper contains 36 sections, 5 equations, 13 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deep Learning Methods.
  4. LLM-Based Methods.
  5. Our Method
  6. Human Expertise Curation
  7. Structured Knowledge Construction.
  8. Verification and Adaptive Refinement.
  9. Architecture of FinHEAR
  10. Adaptive Risk-Aware Decision Making
  11. Experiments
  12. Experimental Setup
  13. Experimental Results for Trend Forecasting Task
  14. Experimental Results for Trading Decision-making Task
  15. Single-Asset Trading Task
  16. ...and 21 more sections

Figures (13)

  • Figure 1: Conceptual motivation behind FinHEAR. The framework is grounded in two behavioral finance principles—information asymmetry and prospect theory—which emphasize the role of expert knowledge and loss-averse risk behavior in financial decision-making.
  • Figure 2: FinHEAR architecture. The system coordinates agents for historical trend analysis, event interpretation, and expert retrieval. These agents are organized in a temporal pipeline with feedback that adjusts past analyses based on outcomes. Risk and direction are predicted separately and combined to generate final trading actions. In the multi-asset setting, actions are dynamically aggregated based on risk-aware signals to determine final allocation positions.
  • Figure 3: Multi-asset portfolio cumulative returns over time for all strategies
  • Figure 4: Portfolio Management Performance Evaluation: FinHEAR vs. Baselines (Test Period)
  • Figure 4: Cumulative trading performance of FinHEAR and other baselines on four individual assets (XAUUSD, AAPL, TSLA, XOM) from January 2023 to December 2023.
  • ...and 8 more figures