Can Large Language Models Execute Parent Orders?

Researchers found that large language models can help execute large stock orders more effectively.

The paper introduces PACE, a hierarchical method that separates long-horizon planning from short-horizon execution.

On Shenzhen Stock Exchange Level-1 data, PACE slightly outperformed TWAP, Almgren-Chriss, and learning-based baselines.

Its behavior differed from human traders, highlighting a distinct execution style from LLMs.

The work expands LLM use in finance from trade selection to cost-efficient order execution.