Stock portfolio selection based on risk appetite: Evidence from ChatGPT

We analyze whether a large language model can generate investment portfolios with varying risk appetites and evaluate their performance against benchmarks. We prompt different ChatGPT models to create portfolios for different risk appetites of retail investors, focusing on U.S. and European equity markets. Our study reveals that higher-risk portfolios yield higher returns. GPT-4o outperforms in the U.S., while GPT-4 offers the highest returns in Europe. We further show that ChatGPT effectively adjusts portfolio risk and returns metrics based on individual risk preferences. These findings suggest private investors can use ChatGPT to improve investment decisions, but careful model selection is vital. • LLM generated portfolios can outperform market benchmarks. • GPT-4o outperforms in U.S. markets, while GPT-4 achieves the highest returns in Europe. • ChatGPT effectively adjusts portfolio risk and return metrics based on investor preferences. • Model selection is crucial for optimizing investment decisions using ChatGPT. • LLMs can reliably reproduce portfolio compositions across repeated prompts.