This paper investigates whether large language models (LLMs) can enhance cross-sectional momentum strategies by extracting predictive signals from firm-specific news. The authors combine daily U.S. equity returns for S&P 500 constituents with high-frequency news data and use prompt-engineered queries to ChatGPT to evaluate whether recent news supports continuation of past returns. The LLM produces scores that condition both stock selection and portfolio weights, creating an LLM-enhanced momentum strategy. The methodology involves a validation set (October 2019 to December 2023) for hyperparameter tuning and a truly out-of-sample test set (January 2024 to March 2025) after ChatGPT's pre-training cutoff, ensuring no information leakage.

The empirical results show that the LLM-enhanced strategy consistently outperforms a standard long-only momentum benchmark, delivering higher Sharpe and Sortino ratios both in-sample and out-of-sample. For example, the out-of-sample Sharpe ratio improves from 0.79 to 1.06, with lower volatility and smaller maximum drawdown. These gains are robust to transaction costs of 2 bps, prompt design, and portfolio constraints. The optimal configuration uses monthly rebalancing, a one-day news lookback, a basic prompt, 50 stocks, value-weighting, and a high weight multiplier. The study finds that the LLM's informational edge is strongest for concentrated, high-conviction portfolios, and that simpler prompts perform slightly better. Overall, the paper demonstrates that LLMs can serve as effective real-time interpreters of financial news, adding incremental value to established factor-based investment strategies.