Researchers release PAWS dataset for policy-driven agentic financial simulations

Researchers introduced PAWS, a Policy-driven Agentic World Simulation dataset designed to connect policy interventions with temporally aligned historical evidence in financial multi-agent simulations. The dataset covers 36 verified U.S. financial and economic policy episodes, 12,727 policy-linked news records, and 65,291 source-grounded stakeholder actions.

Each stakeholder action in PAWS is linked to its supporting news and encoded as a multi-layer event frame that captures interaction mode, financial-action family and subtype, semantic attributes, and conditional mappings to external taxonomies. Entities are resolved to normalized organizations, and actions are aligned with daily market-return context to support policy-agent simulation replay.

In an evaluation on 2,522 stratified action samples, independent AI and human reviewers achieved 89.4% initial agreement on interaction mode, with disagreements subsequently adjudicated by reviewers.

Case studies of the 2008 short-selling ban and the 2001 decimalization recovered documented policy timelines and associated market patterns across both dense and sparse news settings, demonstrating the dataset’s utility in historically grounded analyses.

A replay study further showed that high accuracy in detecting actions can mask failures to detect rare stakeholder actions, identifying action timing and calibration as central challenges for agentic simulations.