Paper: arXiv 2308.00016

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

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.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 6.0/10 (capped from 7.5: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 7.5)

Why this score: The paper introduces a novel interactive alpha mining paradigm and an agentic workflow with specific components (Trading Idea Polisher, Quant Developer, Analyst) and mentions backtesting metrics like IC and Sharpe ratio, demonstrating significant implementation and empirical validation. However, the core methodology focuses on prompt engineering and system architecture using LLMs rather than dense theoretical derivations or advanced mathematics.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Alpha Mining Paradigm Shift"] --> B["Input: Human Quant Ideas<br>+ Financial Data"]
  B --> C["Method: Prompt Engineering<br>with LLMs"]
  C --> D["Process: Interactive<br>Human-AI Co-Creation"]
  D --> E{"Outcome: Alpha-GPT System"}
  E --> F["Output: Creative,<br>Insightful Alphas"]
  F --> G["Result: Improved<br>Investment Signals"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes Alpha-GPT, an interactive alpha mining system that uses human-AI interaction and large language models to generate creative and effective trading signals.

Source sentence
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.

Assumes

No explicit assumptions found in the text.

Data

  • WorldQuant International Quant Championship (IQC) 2024 · WorldQuant alternative unknown
    Source sentence
    ... The platform applies strict criteria for alpha qualification, considering factors such as alpha return, turnover, and Sharpe ratio.

Costs

Not modeled A trading or execution result is reported without costs.

Finds 2

  • Alpha-GPT consistently improves factor information coefficient (IC) over multiple search enhancement iterations, with both in-sample and out-of-sample IC showing sharp initial increases and convergence around the 15th iteration. performance
    Source sentence
    Figure 4 illustrates the IC curves over 20 iterations, revealing several key insights. Both in-sample and out-of-sample ICs show a sharp initial increase (iterations 0 to 5)
  • In the WorldQuant IQC competition, Alpha-GPT generated alphas that ranked among the top 10 worldwide and top 3 regionally, achieving the highest in-sample score and competitive out-of-sample performance. performance
    Source sentence
    the results demonstrate that our automated Alpha-GPT system can generate performant alphas, ranking among the top 10 worldwide and top 3 regionally. In particular, Alpha-GPT produces a comparable number of qualified alphas to top human competitors and achieves the highest in-sample score.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 7 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 28 papers in the archive (23 by independent authors): Large Language Model Agent in Financial Trading: A Survey (discusses its findings); AlphaRLM: Aligning Language Models for Literature-Grounded … (uses its method or data); To Trade or Not to Trade: An Agentic Approach to Estimating … (uses its method or data)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.