Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Design and Empirical Study of a Large Language Model-Based Multi-Agent Investment System for Chinese Public REITs

Zheng Li

TL;DR

This work tackles the low-volatility Chinese public REITs market by proposing an LLM-driven multi-agent trading framework that integrates four analytical agents with a fusion-based prediction and a discrete execution loop. It compares two prediction pathways, DeepSeek-R1 and a fine-tuned Qwen3-8B, demonstrating significant outperformance versus Buy & Hold in a 12-month backtest (Oct 2024–Oct 2025) in terms of cumulative return and risk-adjusted metrics. Key contributions include structured interpretation across information sources, multi-horizon directional predictions, and closed-loop execution with risk controls, as well as a rigorous two-stage fine-tuning approach (SFT + GSPO) for the smaller LLM. The results indicate that a multi-agent, LLM-enhanced framework can improve REITs trading performance and risk management, with the fine-tuned small model achieving comparable or superior performance in some scenarios, and the work provides open-source code and data for replication.

Abstract

This study addresses the low-volatility Chinese Public Real Estate Investment Trusts (REITs) market, proposing a large language model (LLM)-driven trading framework based on multi-agent collaboration. The system constructs four types of analytical agents-announcement, event, price momentum, and market-each conducting analysis from different dimensions; then the prediction agent integrates these multi-source signals to output directional probability distributions across multiple time horizons, then the decision agent generates discrete position adjustment signals based on the prediction results and risk control constraints, thereby forming a closed loop of analysis-prediction-decision-execution. This study further compares two prediction model pathways: for the prediction agent, directly calling the general-purpose large model DeepSeek-R1 versus using a specialized small model Qwen3-8B fine-tuned via supervised fine-tuning and reinforcement learning alignment. In the backtest from October 2024 to October 2025, both agent-based strategies significantly outperformed the buy-and-hold benchmark in terms of cumulative return, Sharpe ratio, and maximum drawdown. The results indicate that the multi-agent framework can effectively enhance the risk-adjusted return of REITs trading, and the fine-tuned small model performs close to or even better than the general-purpose large model in some scenarios.

Design and Empirical Study of a Large Language Model-Based Multi-Agent Investment System for Chinese Public REITs

TL;DR

This work tackles the low-volatility Chinese public REITs market by proposing an LLM-driven multi-agent trading framework that integrates four analytical agents with a fusion-based prediction and a discrete execution loop. It compares two prediction pathways, DeepSeek-R1 and a fine-tuned Qwen3-8B, demonstrating significant outperformance versus Buy & Hold in a 12-month backtest (Oct 2024–Oct 2025) in terms of cumulative return and risk-adjusted metrics. Key contributions include structured interpretation across information sources, multi-horizon directional predictions, and closed-loop execution with risk controls, as well as a rigorous two-stage fine-tuning approach (SFT + GSPO) for the smaller LLM. The results indicate that a multi-agent, LLM-enhanced framework can improve REITs trading performance and risk management, with the fine-tuned small model achieving comparable or superior performance in some scenarios, and the work provides open-source code and data for replication.

Abstract

This study addresses the low-volatility Chinese Public Real Estate Investment Trusts (REITs) market, proposing a large language model (LLM)-driven trading framework based on multi-agent collaboration. The system constructs four types of analytical agents-announcement, event, price momentum, and market-each conducting analysis from different dimensions; then the prediction agent integrates these multi-source signals to output directional probability distributions across multiple time horizons, then the decision agent generates discrete position adjustment signals based on the prediction results and risk control constraints, thereby forming a closed loop of analysis-prediction-decision-execution. This study further compares two prediction model pathways: for the prediction agent, directly calling the general-purpose large model DeepSeek-R1 versus using a specialized small model Qwen3-8B fine-tuned via supervised fine-tuning and reinforcement learning alignment. In the backtest from October 2024 to October 2025, both agent-based strategies significantly outperformed the buy-and-hold benchmark in terms of cumulative return, Sharpe ratio, and maximum drawdown. The results indicate that the multi-agent framework can effectively enhance the risk-adjusted return of REITs trading, and the fine-tuned small model performs close to or even better than the general-purpose large model in some scenarios.
Paper Structure (31 sections, 14 equations, 2 figures)

This paper contains 31 sections, 14 equations, 2 figures.

Table of Contents

  1. Introduction
  2. Structured Trading Systems with Multi-Agents
  3. Fine-tuning and Reinforcement Learning Post-Training
  4. Positioning and Contributions of This Study
  5. System Architecture
  6. Price Momentum Agent
  7. Technical Indicator Calculation and Technical State Construction
  8. Dynamic Volatility Threshold and "Sideways" Modeling
  9. LLM-Driven Technical Signal Synthesis
  10. Announcement Agent
  11. Event Agent
  12. Market Agent
  13. Multi-layer Market State Construction
  14. Four-Quadrant Macro Allocation Framework (Core)
  15. LLM-Driven Comprehensive Market Interpretation
  16. ...and 16 more sections

Figures (2)

  • Figure 1: Multi-Agent REITs Trading System Architecture Diagram
  • Figure 2: Comparison of Aggregate Total NAV across All Fund Accounts for the Three Strategies