Training Language Models for Bilateral Trade with Private Information
Dirk Bergemann, Soheil Ghili, Xinyang Hu, Zhuoran Yang · 2026
Read Training Language Models for Bilateral Trade with Private Information by Dirk Bergemann, Soheil Ghili, Xinyang Hu, Zhuoran Yang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-16472.txt.
Description
Bergemann and colleagues test language models as bargainers in trade with private information, where rationality, surplus and cooperation matter. Their event-driven simulator separates binding offers from messages, serving as benchmark and training ground. A round-robin of five frontier models over 15,000 negotiations found patience, calibrated concessions, aggressive anchoring and price discrimination through sequential offers won the most surplus and deals. Training Qwen3 showed a trade-off: supervised tuning roughly doubled surplus share but cut deal rates, while reinforcement learning restored deals at some surplus cost.
Catalog record
| CREATOR | Dirk Bergemann, Soheil Ghili, Xinyang Hu, Zhuoran Yang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 124226 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.16472