ChatGPT in Systematic Investing: Enhancing Risk-Adjusted Returns with LLMs

Battle Count: 272

Highly relevant to quantitative trading. The paper directly addresses enhancing a well-known factor strategy (cross-sectional momentum) with LLM-generated signals from real-time news. It provides concrete performance improvements (Sharpe ratio from 0.79 to 1.06 out-of-sample), discusses transaction costs, portfolio constraints, rebalancing frequency, and practical implementation considerations. The methodology is directly applicable to systematic equity strategies and demonstrates how LLMs can serve as real-time interpreters of financial news for alpha generation.

Moderate complexity. The core methodology involves: (1) computing standard momentum rankings (simple), (2) collecting stock-specific news from an API (moderate), (3) constructing and sending prompts to ChatGPT API for each stock-day (moderate, requires API management and rate limiting), (4) normalizing scores and re-ranking stocks (simple), (5) portfolio construction with weight tilting (simple). The main challenges are: managing API costs and latency for daily processing of many stocks, handling missing news, ensuring consistent prompt formatting, and the hyperparameter optimization grid (512 combinations). No model training or fine-tuning is required.

The paper uses proprietary internal Finreon data (similar to CRSP), a commercial Stock News API, and ChatGPT 4.0 mini via OpenAI API. No GitHub repository is provided. The prompt templates are disclosed in the paper, and the methodology is described in detail, but the core datasets are not publicly available. The LLM is a closed-source commercial model. Reproduction would require access to the same news API and equity return data, plus API access to ChatGPT 4.0 mini.