Scholarlink

The emergence of Large Language Models (LLMs), has created novel opportunities for advancing portfolio optimization models. This study proposes a method to integrate ChatGPT-4 into existing portfolio construction frameworks to enhance their effectiveness on portfolio management. Specifically, we leverage fundamental factors in the Gated Recurrent Unit (GRU) model and the Chinese version of the Fama-French three-factor model (CH-3) to construct the initial portfolio, and further enhance the model through text analysis reweighting and prompt engineering. Our findings demonstrate that portfolios optimized through ChatGPT-based shareholding tuning consistently achieve superior average returns across various scoring metrics and stock selection models. The findings contribute to both theoretical understanding and practical implementation of AI-enhanced portfolio management models.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Standing on the Shoulder of Giants: Integrating Generative Large Language Models into Portfolio Optimization

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zhenyang Xin,
  • Yue Guan,
  • Xiaoling Hao,
  • Donghan Wang

摘要

The emergence of Large Language Models (LLMs), has created novel opportunities for advancing portfolio optimization models. This study proposes a method to integrate ChatGPT-4 into existing portfolio construction frameworks to enhance their effectiveness on portfolio management. Specifically, we leverage fundamental factors in the Gated Recurrent Unit (GRU) model and the Chinese version of the Fama-French three-factor model (CH-3) to construct the initial portfolio, and further enhance the model through text analysis reweighting and prompt engineering. Our findings demonstrate that portfolios optimized through ChatGPT-based shareholding tuning consistently achieve superior average returns across various scoring metrics and stock selection models. The findings contribute to both theoretical understanding and practical implementation of AI-enhanced portfolio management models.