Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automate Strategy Finding with LLM in Quant Investment

Zhizhuo Kou, Holam Yu, Junyu Luo, Jingshu Peng, Xujia Li, Chengzhong Liu, Juntao Dai, Lei Chen, Sirui Han, Yike Guo

TL;DR

This work addresses brittleness in traditional quantitative finance models by introducing a three-stage framework that leverages large language models (LLMs) within a risk-aware multi-agent system to automate alpha factor discovery and strategy optimization. It combines an LLM-based Seed Alpha Factory, a multimodal multi-agent evaluation (Confidence Score Agent and Risk Preference Agent), and a dynamic weight-optimization module (a 3-layer MLP) to produce adaptive trading signals across diverse markets. Empirical results in the SSE50, CSI300, and SP500 demonstrate robust alpha generation and superior risk-adjusted performance, including a 53.17% cumulative return in 2023 on SSE50 and strong cross-market resilience. The approach extends Fin-LLMs and multi-agent frameworks to scalable, automated portfolio construction with practical implications for real-world quantitative trading.

Abstract

We present a novel three-stage framework leveraging Large Language Models (LLMs) within a risk-aware multi-agent system for automate strategy finding in quantitative finance. Our approach addresses the brittleness of traditional deep learning models in financial applications by: employing prompt-engineered LLMs to generate executable alpha factor candidates across diverse financial data, implementing multimodal agent-based evaluation that filters factors based on market status, predictive quality while maintaining category balance, and deploying dynamic weight optimization that adapts to market conditions. Experimental results demonstrate the robust performance of the strategy in Chinese & US market regimes compared to established benchmarks. Our work extends LLMs capabilities to quantitative trading, providing a scalable architecture for financial signal extraction and portfolio construction. The overall framework significantly outperforms all benchmarks with 53.17% cumulative return on SSE50 (Jan 2023 to Jan 2024), demonstrating superior risk-adjusted performance and downside protection on the market.

Automate Strategy Finding with LLM in Quant Investment

TL;DR

This work addresses brittleness in traditional quantitative finance models by introducing a three-stage framework that leverages large language models (LLMs) within a risk-aware multi-agent system to automate alpha factor discovery and strategy optimization. It combines an LLM-based Seed Alpha Factory, a multimodal multi-agent evaluation (Confidence Score Agent and Risk Preference Agent), and a dynamic weight-optimization module (a 3-layer MLP) to produce adaptive trading signals across diverse markets. Empirical results in the SSE50, CSI300, and SP500 demonstrate robust alpha generation and superior risk-adjusted performance, including a 53.17% cumulative return in 2023 on SSE50 and strong cross-market resilience. The approach extends Fin-LLMs and multi-agent frameworks to scalable, automated portfolio construction with practical implications for real-world quantitative trading.

Abstract

We present a novel three-stage framework leveraging Large Language Models (LLMs) within a risk-aware multi-agent system for automate strategy finding in quantitative finance. Our approach addresses the brittleness of traditional deep learning models in financial applications by: employing prompt-engineered LLMs to generate executable alpha factor candidates across diverse financial data, implementing multimodal agent-based evaluation that filters factors based on market status, predictive quality while maintaining category balance, and deploying dynamic weight optimization that adapts to market conditions. Experimental results demonstrate the robust performance of the strategy in Chinese & US market regimes compared to established benchmarks. Our work extends LLMs capabilities to quantitative trading, providing a scalable architecture for financial signal extraction and portfolio construction. The overall framework significantly outperforms all benchmarks with 53.17% cumulative return on SSE50 (Jan 2023 to Jan 2024), demonstrating superior risk-adjusted performance and downside protection on the market.
Paper Structure (31 sections, 9 equations, 4 figures, 11 tables, 2 algorithms)

This paper contains 31 sections, 9 equations, 4 figures, 11 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Alpha Factor Representation
  4. Alpha Mining and Selection
  5. Strategy Optimization
  6. Methodology
  7. Framework Overview
  8. LLM-Based Seed Alpha Generation
  9. Multimodal Multi-Agent Evaluation
  10. Weight Optimization
  11. Experiment
  12. Datasets
  13. Multimodal Knowledge Extraction and Adaptive Alpha Discovery
  14. Comparative Performance Against Traditional Alpha Factories
  15. Performance Evaluation of the Integrated Alpha Framework in Benchmark Comparison
  16. ...and 16 more sections

Figures (4)

  • Figure 1: Overview of the strategy generate process in three components, a Seed Alpha Factory built using LLMs, a multi-agent decision-making system, and a weight optimization approach for overall strategy (CS stands for confidence score; RP stands for risk preference)
  • Figure 2: Sample Experiment on Different Market Status Input and Alpha Selection, Selected Alpha Depends on Different Context
  • Figure 3: Cumulative return backtest result on SSE50. The line track the net worth of different methods
  • Figure 4: Seed Alpha Representation: (A) An example of the seed alpha formula. (B) Its equivalent expression tree. (C) Step-by-step computation of this seed alpha on an example time series.