Directed Herding in Financial Markets
A reproducible empirical research program and live research instrument that integrates quarterly SEC Form 13F institutional holdings, official information events, and a Wikimedia Pageviews retail-attention proxy to surface transparent, non-causal candidate herding patterns for review and empirical testing.
Research Question
Can synchronized changes in institutional portfolios be distinguished from independent reactions to common public information, mechanical rebalancing, and other non-herding explanations?
Why This Matters
The main contribution of this work is creating a traceable path from public disclosures and attention signals to reviewable hypotheses, without treating synchronization as proof of imitation, coordination, or causation. Making that path explicit is a prerequisite for any credible empirical claim about herding.
Current Direction
Operating a governed ten-manager research sample with automated weekly polling for new and amended SEC 13F filings (while the underlying holdings remain quarterly), official Federal Reserve and BLS event capture, and a Wikimedia Pageviews attention proxy. The platform layers portfolio-overlap, synchronized-change, screening, hypothesis, workflow, and analysis-ready feature views on top of that base. Ongoing work extends historical depth, event and attention coverage, metadata quality, and formal statistical testing.
Early Notes
The platform already produces descriptive candidate episodes, participant records, mechanical-risk flags, literature-review tasks, experiment plans, and daily research briefs. All outputs remain exploratory until identifiers and sectors, event context, retail context, additional quarters, and identification tests are sufficiently complete.
Research Materials
Documents the architecture, sources, automation, sampling, methods, governance, and limitations of the Directed Herding Research Instrument.
Source repository for the research platform frontend and Cloudflare deployment.