Background
Latest research

Research on Latent Analyst Herding

Original research, white papers, and market commentary on analyst herding and behavioral market dynamics. We publish what we find, even when it contradicts what we expected.

Open methodology

We publish our model specifications, estimation procedures, and replication materials so others can scrutinize, extend, or challenge our work.

Full disclosure

When our models fail, we report the failure. When our predictions miss, we analyze why. Intellectual honesty is not a slogan here.

Peer collaboration

Our research improves when others push back. We actively seek external review, methodological critique, and adversarial testing of our claims.

Featured Research: The Quiet Pull of Consensus

Our most recent paper examines a herding event in technology sector earnings forecasts during the second quarter of 2025. The convergence was invisible to standard consensus metrics. Our latent framework detected it three weeks before the visible cluster formed, and traced it to a specific information pathway between two brokerages.

analyst network convergence visualization

Detection methodology

We applied a spatial dependence model to a panel of forty-seven analyst estimates, extracting a latent correlation matrix that revealed a leader-follower structure operating beneath the surface consensus. The model identified the initiating revision and mapped the subsequent cascade through the analyst network.

market price reaction timeline

Market impact analysis

The herding cluster produced a predictable post-earnings drift. Forecasts within the cluster were systematically optimistic relative to actual earnings, and the market corrected gradually over the subsequent eight trading sessions as independent information filtered through.

statistical validation process

Validation protocol

We subjected the detected regime to placebo simulations, alternative specifications, and out-of-sample testing across four subsequent quarters. The herding signal persisted in three of four tests, with effect sizes consistent with the original detection.

Research Milestones

A chronological record of how our framework for detecting latent analyst herding has evolved from initial curiosity to practical application across institutional research settings.

2020

The Initial Question

Dr. Eleanor Vance-Chen observed that analyst forecast revisions in the technology sector were clustering in ways that standard information models could not explain. She began developing a latent variable framework to measure hidden dependence structures in estimate panels. The first prototype was a simple spatial autoregressive model applied to a single quarter of earnings data.

Origin Prototype
2021

Methodology Publication

The first formal paper on latent herding detection was presented at the Sydney Finance Symposium. The methodology introduced a three-component decomposition of forecast correlation into common factors, idiosyncratic errors, and latent spatial terms. Peer feedback led to significant refinements in the estimation procedure.

Publication Peer review
2022

Open-Source Release

We released the first version of our herding detection toolkit under an open license. The codebase included data preparation scripts, model estimation routines, and validation testing modules. External researchers began replicating and extending the methodology across different sectors and geographies.

Open source Community
2023

Regime Classification Framework

Building on two years of empirical results, we introduced a taxonomy of herding mechanisms. The framework distinguished leader-follower structures, mutual influence clusters, and information cascades, each with distinct implications for forecast quality and market impact. This became the basis for our current alert classification system.

Taxonomy Classification
2024

Institutional Integration

Three quantitative research teams adopted the latent herding framework into their earnings surprise models. We collaborated on implementation, provided methodological support, and gathered feedback that led to improvements in computational efficiency and real-time signal generation. The framework moved from research prototype to production tool.

Adoption Production
2025

Cross-Sector Expansion

We extended the methodology from technology and healthcare into energy, consumer discretionary, and financial services. Each sector presented distinct herding dynamics, requiring sector-specific calibration of the latent variable models. The expanded dataset now covers over two hundred analyst-firm combinations across five major sectors.

Expansion Calibration
Six years of development

Recent developments in our ongoing study of latent herding patterns across global markets, including new dataset releases, methodological refinements, and conference presentations.

Research Updates

Paper

Spatial Dependence in Emerging Market Forecasts

New research extends the latent herding framework to analyst coverage of emerging market equities, where information asymmetry and brokerage concentration create distinct herding dynamics.

emerging market research visualization
Methodology

Bayesian Hierarchical Models for Herding Detection

We introduce a Bayesian estimation approach that improves identification of herding regimes in small-sample panels, addressing a key limitation of maximum likelihood methods.

Event

Presentation at Melbourne Quantitative Forum

Our team presented findings on regime-switching herding patterns in Australian equity markets, with discussion of practical applications for local institutional investors.

Data

Q3 2025 Herding Regime Dataset Released

The quarterly update to our herding detection database includes newly classified regimes, expanded analyst coverage, and revised baseline estimates for four major sectors.

Paper

Critique of Standard Consensus Metrics

A working paper argues that conventional dispersion measures systematically miss latent herding by conflating common information shocks with behavioral convergence.

Event

Workshop on Real-Time Herding Detection

We hosted a practical workshop for institutional researchers on implementing latent herding models in production environments, with live coding and dataset walkthroughs.