Large Language Models in Systematic Investing
ChatGPT systematic investing is an emerging area of research exploring how large language models can improve quantitative investment strategies. This paper investigates whether models such as ChatGPT can enhance cross-sectional momentum strategies by extracting predictive signals from firm-specific financial news.
Traditional factor strategies rely primarily on price and fundamental data. In contrast, LLMs can interpret unstructured information such as news articles, enabling new sources of predictive signals for quantitative investment models.
ChatGPT Systematic Investing and Momentum Strategies
The study investigates whether LLMs can improve cross-sectional momentum strategies. Using daily U.S. equity returns for S&P 500 constituents combined with high-frequency news data, we generate signals that determine whether recent news supports the continuation of past returns.
Prompt-engineered queries are sent to ChatGPT whenever a stock is about to enter a momentum portfolio. The LLM evaluates the tone and relevance of recent news events and produces scores that influence both stock selection and portfolio weighting.
Data and Methodology
The dataset combines daily returns for S&P 500 stocks with a stream of firm-specific news data. The LLM analyzes these news events and determines whether they reinforce or contradict the stock’s momentum signal.
By integrating this information, the model adjusts the composition of the momentum portfolio in real time. This approach effectively blends traditional factor investing with natural language processing techniques.
Performance of the LLM Strategy
The LLM-enhanced momentum strategy outperforms a traditional long-only momentum benchmark. The strategy delivers higher Sharpe ratios and Sortino ratios both in-sample and in a truly out-of-sample period that occurs after the LLM’s training cutoff.
Importantly, the improvement in performance remains robust after accounting for:
- transaction costs
- alternative prompt designs
- portfolio constraints
The gains are most pronounced in concentrated, high-conviction portfolios, where the informational advantage of the LLM signal becomes more impactful.
Key Findings
- Large language models can extract predictive signals from financial news.
- LLM-generated signals improve risk-adjusted returns in momentum strategies.
- Performance gains remain robust across different prompts and trading constraints.
- LLMs can act as real-time interpreters of financial information within systematic investment frameworks.
Implications for Quantitative Investing
The findings suggest that LLMs can complement traditional factor models by adding information derived from unstructured data sources such as news articles.
Rather than replacing systematic strategies, LLMs may serve as an additional layer of alpha generation, improving portfolio construction and signal validation in quantitative investment systems.