Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·27d ago·5 min read
Are LLMs Good Financial User Simulators? A Preliminary Study
TL;DR
A preliminary study evaluates the effectiveness of large language models (LLMs) as financial user simulators in predicting trading behaviors. Results indicate that while market context enhances predictions, LLMs struggle with transaction sizing and exhibit systematic behavioral biases.
✦ Why It Matters
Consider testing LLMs in your financial simulations to evaluate their predictive capabilities and biases.
Key Takeaways
- Test LLMs for predicting trading actions in your applications.
- Evaluate the impact of market context on prediction accuracy.
- Be aware of LLM biases in trading simulations.
- Consider larger-scale studies for improved behavioral fidelity.
How It Works
Participants engaged in paper trading with virtual funds, allowing researchers to analyze LLM predictions based on historical market data. The LLMs were tasked with predicting the next trading day's actions, securities, and transaction sizes, using only information available before a specific cutoff.