LLMs Enhance Sector Portfolio Construction With Caveats

Researchers evaluate LLMs from OpenAI, Google, Anthropic, DeepSeek, and xAI for S&P 500 sector portfolio construction, prompting each model to select and weight 20 stocks per sector and testing performance across Jan–Mar 2025 (stable) and Apr–Jun 2025 (volatile). They find LLM portfolios often outperform in stable markets but underperform in volatile regimes; combining LLM selection with classical optimization improves performance and consistency.