Automate Strategy Finding with LLM in Quant Investment (2409.06289v3)

Abstract: We present a novel three-stage framework leveraging 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.

• The paper introduces a novel framework using Large Language Models (LLMs) and a multi-agent system to automate finding and optimizing quantitative stock investment strategies from multimodal financial data, addressing the instability of existing deep learning models.
• The framework involves LLMs generating potential alpha factors from diverse data sources, which are then evaluated dynamically by a multi-agent system considering different risk preferences and market conditions.
• Validated on Chinese stock markets, the approach demonstrated substantial outperformance and improved stability compared to baseline methods, utilizing a dynamic weighting mechanism for adaptive strategy optimization based on real-time data.

The paper "Automate Strategy Finding with LLM in Quant Investment" introduces an advanced framework for quantitative stock investment, particularly in portfolio management and alpha mining, leveraging the capabilities of LLMs and multi-agent systems. The primary motivation for this research is the observed instability and uncertainty in existing deep learning models used in financial trading which hinder their practical application.

Framework Overview:

1. Seed Alphas Generation with LLMs: The proposed framework begins by utilizing LLMs to mine alpha factors from multimodal financial data. This includes data not only from numerical sources but also from research papers, visual charts, and other non-standard data types relevant for comprehensive market analysis. The purpose is to generate a wide variety of predictive signals that go beyond traditional heuristic rules.
2. Multi-Agent System for Alpha Evaluation: A multi-agent system is employed to harness these predictive signals. Multiple agents, each with different risk preferences, are tasked with evaluating the potential alpha factors. This dynamic evaluation allows strategies to adapt to a fluid market environment, selecting alphas that align well with current market conditions and ensuring robustness against various financial scenarios.
3. Dynamic Weighting and Strategy Optimization: The selected alpha factors are subjected to a weight-gating mechanism, which assigns optimal weights based on real-time market data. This process creates a customized composite alpha formula that is highly context-aware and capable of adapting to significant market changes.

Implementation and Results:

The framework was experimentally validated on the Chinese stock markets, where it demonstrated substantial outperformance compared to state-of-the-art baseline methods across several financial metrics. Key metrics used in validating the model included the Information Coefficient (IC) and other traditional financial performance indicators. The results indicated that the integration of LLM-generated alphas and a multi-agent architecture yielded superior trading performance and improved stability, suggesting that AI-driven approaches have significant potential to enhance quantitative investment strategies.

Significant Contributions:

• The paper introduces a novel integration of LLMs in financial trading, highlighting their ability to mine extensive and diversified alpha factors from a comprehensive set of multimodal data sources.
• It leverages a multi-agent system to adaptively evaluate and select alpha factors based on varying market conditions, effectively managing market variability.
• The proposed dynamic strategy optimization approach, through the application of deep learning techniques, ensures that the derived strategies are both robust and adaptable, ultimately setting a new benchmark for machine learning application in financial trading.

In conclusion, this research advances the field of quantitative trading by offering a robust, adaptive framework that utilizes cutting-edge AI tools to optimize investment strategies across various asset classes, thus offering a versatile solution applicable to global financial markets. The proposed approach underscores the growing potential of AI in the field of quantitative finance, setting a precedent for future research in integrating multimodal data and machine learning in stock trading strategies.