RESEARCH · RESEARCH · #1023
PAWS: policy-driven agentic world simulation dataset for U.S. financial episodes (arXiv:2609.28547v1)
PAWS is a new dataset for policy-driven multi-agent financial simulation covering 36 verified U.S. policy episodes, 12,727 policy-linked news records, and 65,291 source-grounded stakeholder actions, each linked to supporting news and daily market-return context. The release includes multi-layer event frames, entity normalization, reviewer adjudication (89.4% initial agreement on 2,522 stratified action samples), case studies recovering timelines for the 2008 short-selling ban and 2001 decimalization, and replay analyses highlighting rare-action detection and timing/calibration challenges.
KEY POINTS
- PAWS is a new dataset for policy-driven multi-agent financial simulation covering 36 verified U.S.
- policy episodes, 12,727 policy-linked news records, and 65,291 source-grounded stakeholder actions, each linked to supporting news and daily market-return context.
- The release includes multi-layer event frames, entity normalization, reviewer adjudication (89.4% initial agreement on 2,522 stratified action samples), case studies recovering timelines for the 2008 short-selling ban and 2001 decimalization, and replay analyses highlighting rare-action detection and timing/calibration challenges.
WHY IT MATTERS
PAWS creates an auditable, historically grounded substrate to evaluate agent influence, policy–response cascades, and action–outcome alignment in financial multi-agent simulations.