本研究探讨了大型语言模型在债券市场因子生成中的应用潜力，具体研究了OpenAI开发的ChatGPT-4模型能否只利用基础价格指标和债券收益率的数据结构名称而不依赖具体的数值构造出能带来高收益的债券因子。本研究依托WRDS数据库中2012年至2023年的公司债券日度数据，引导ChatGPT生成了30个符合代码正确性和一定复杂性要求的因子。

实证结果显示，ChatGPT生成的因子能够在日换仓下达到1.5的年化收益率，即使在损失信息进行周换仓的情况下，许多因子也能保持20\%以上的年化收益率。通过在不同的时间测试集上进行检验，本研究发现GPT生成的债券因子在不同的时间段都能维持正的高收益和高夏普比率，而不只是局限于特定的时间点。通过构建一个简单的均值多因子模型，本研究进一步验证了这些因子的有效性。债券按因子均值分组的投资组合表现出因子均值与年化收益率和夏普比率的正相关性，与最大回撤率的负相关性，证实了这些因子的市场预测能力。

研究突出了大型语言模型在金融领域的潜力，扩展了大语言模型在债券资产定价研究领域的应用，促进金融分析方法论的创新，为未来金融研究提供新的工具和框架，并为传统的债券因子构造方法提供了一种高效率且可行性高的替代方案。

This study explores the potential of large language models for bond factor generation, specifically examining the ability of OpenAI's ChatGPT-4 model to construct high-yield bond factors solely using basic price indicators and bond yield data structure names without relying on specific values. Utilizing American daily corporate bond data from 2012 to 2023, the study guides ChatGPT to generate 30 factors that met the criteria for code correctness and complexity.

Empirical results show that some factors generated by ChatGPT can achieve an annualized return rate of 1.5 under daily rebalancing, and many maintain an annualized return rate of over 20\% even under weekly rebalancing, where information loss occurs. By testing on different time datasets, this study find that GPT-generated bond factors maintained positive high returns and high Sharpe ratios over various periods, not just at specific points in time. By constructing a simple mean multi-factor model, this study further validated the effectiveness of these factors. Investment portfolios grouped by factor mean demonstrated a positive correlation between factor mean and both annualized return rate and Sharpe ratio, and a negative correlation with maximum drawdown rate, confirming the market prediction capabilities of these factors.

The research highlights the potential of large language models in the financial sector, extends the application of large language models in the field of bond asset pricing research, promotes innovation in financial analysis methodology, provides new tools and frameworks for future financial research, and offers a highly efficient and feasible alternative to traditional methods of bond factor construction.