Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Alpha-GPT: Human-AI Interactive Alpha Mining for Quantitative Investment

Saizhuo Wang, Hang Yuan, Leon Zhou, Lionel M. Ni, Heung-Yeung Shum, Jian Guo

TL;DR

The paper tackles the bottleneck in quantitative alpha mining by introducing a human-AI interaction paradigm embodied in Alpha-GPT, which couples a tri-stage agentic workflow (ideation, implementation, review) with two operation modes (interactive and autonomous). It details an end-to-end architecture consisting of a WebUI/AlphaBot, a GP-enhanced algorithmic mining backend, and a computation-acceleration layer, anchored by a knowledge library for grounding and interpretability. Through quantitative and qualitative evaluations, including translation-consistency tests, iterative refinement IC gains, and strong results in the HF competition and WorldQuant IQC 2024, the work demonstrates improved efficiency and the generation of high-quality, explainable alphas. The findings suggest that LLM-mediated human-AI collaboration can accelerate alpha discovery while maintaining generalization and interpretability, offering a practical path for scalable, creative quantitative research.

Abstract

One of the most important tasks in quantitative investment research is mining new alphas (effective trading signals or factors). Traditional alpha mining methods, either hand-crafted factor synthesizing or algorithmic factor mining (e.g., search with genetic programming), have inherent limitations, especially in implementing the ideas of quants. In this work, we propose a new alpha mining paradigm by introducing human-AI interaction, and a novel prompt engineering algorithmic framework to implement this paradigm by leveraging the power of large language models. Moreover, we develop Alpha-GPT, a new interactive alpha mining system framework that provides a heuristic way to ``understand'' the ideas of quant researchers and outputs creative, insightful, and effective alphas. We demonstrate the effectiveness and advantage of Alpha-GPT via a number of alpha mining experiments.

Alpha-GPT: Human-AI Interactive Alpha Mining for Quantitative Investment

TL;DR

The paper tackles the bottleneck in quantitative alpha mining by introducing a human-AI interaction paradigm embodied in Alpha-GPT, which couples a tri-stage agentic workflow (ideation, implementation, review) with two operation modes (interactive and autonomous). It details an end-to-end architecture consisting of a WebUI/AlphaBot, a GP-enhanced algorithmic mining backend, and a computation-acceleration layer, anchored by a knowledge library for grounding and interpretability. Through quantitative and qualitative evaluations, including translation-consistency tests, iterative refinement IC gains, and strong results in the HF competition and WorldQuant IQC 2024, the work demonstrates improved efficiency and the generation of high-quality, explainable alphas. The findings suggest that LLM-mediated human-AI collaboration can accelerate alpha discovery while maintaining generalization and interpretability, offering a practical path for scalable, creative quantitative research.

Abstract

One of the most important tasks in quantitative investment research is mining new alphas (effective trading signals or factors). Traditional alpha mining methods, either hand-crafted factor synthesizing or algorithmic factor mining (e.g., search with genetic programming), have inherent limitations, especially in implementing the ideas of quants. In this work, we propose a new alpha mining paradigm by introducing human-AI interaction, and a novel prompt engineering algorithmic framework to implement this paradigm by leveraging the power of large language models. Moreover, we develop Alpha-GPT, a new interactive alpha mining system framework that provides a heuristic way to ``understand'' the ideas of quant researchers and outputs creative, insightful, and effective alphas. We demonstrate the effectiveness and advantage of Alpha-GPT via a number of alpha mining experiments.
Paper Structure (28 sections, 8 figures, 5 tables)

This paper contains 28 sections, 8 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Agentic Workflow
  3. Ideation
  4. Implementation
  5. Review
  6. Modes of Operation
  7. Interactive Mode
  8. Autonomous Mode
  9. System Architecture
  10. WebUI and LLM Agent
  11. Backend Systems
  12. Algorithmic Alpha Mining
  13. Alpha Computation Acceleration
  14. Evaluations
  15. Data and operators
  16. ...and 13 more sections

Figures (8)

  • Figure 1: Evolution of alpha mining techniques.
  • Figure 2: The agentic workflow of Alpha-GPT
  • Figure 3: Alpha-GPT's hierarhical RAG in autonomous mode for large-scale quant database.
  • Figure 4: Search Enhancement curve
  • Figure 5: Alpha-GPT internal working pipeline: After a user inputs their ideas, the system goes into the knowledge compilation module. It uses external memory to pull similar examples, and combines them into the system prompt. The module passes everything to the LLM which creates valid alpha expressions and config files. These alphas are evaluated via Alpha Search, and results are presented to the user along with an interpretation provided by the Thoughts Decompiler.
  • ...and 3 more figures